# Shravonix Documentation > Complete documentation and content for AI understanding *Complete documentation content below* # Tech Deep Dives > In-depth technical analysis on AI models, browser engines, web frameworks, Linux kernel, and hardware. Nerdy insights for developers and tech enthusiasts who want deep technical content. // Latest [// ai](/articles/ai-in-india-2026-scared-leading-and-wasting-water) ## [AI in India 2026: Scared, Leading, and Wasting Water](/articles/ai-in-india-2026-scared-leading-and-wasting-water) [India leads the world in enterprise AI at 80% but ranks 101st per person. The full honest picture: who's winning, what it costs, and what you should actually do.](/articles/ai-in-india-2026-scared-leading-and-wasting-water) [Jainil Prajapati ](/authors/jainil-prajapati)2026-05-16 [// ai](/articles/when-claude-became-my-relationship-coach-she-hated-ai) ### [When Claude Became My Relationship Coach (She Hated AI)](/articles/when-claude-became-my-relationship-coach-she-hated-ai) [2026-05-11](/articles/when-claude-became-my-relationship-coach-she-hated-ai) [// ai](/articles/tokens-are-getting-cheaper-so-why-does-ai-feel-more-expensive) ### [Tokens Are Getting Cheaper. So Why Does AI Feel More Expensive?](/articles/tokens-are-getting-cheaper-so-why-does-ai-feel-more-expensive) [2026-05-08](/articles/tokens-are-getting-cheaper-so-why-does-ai-feel-more-expensive) // More All Linux Tech AI IoT Opinions Programming SEARCH ARTICLES... [// ai](/articles/tokens-are-cheap-thinking-isnt) ### [Tokens Are Cheap. Thinking Isn't.](/articles/tokens-are-cheap-thinking-isnt) [2026-04-24](/articles/tokens-are-cheap-thinking-isnt) [// ai](/articles/india-was-the-worlds-ai-warfare-lab-heres-what-actually-happened) ### [India Was the World's AI Warfare Lab. Here's What Actually Happened.](/articles/india-was-the-worlds-ai-warfare-lab-heres-what-actually-happened) [2026-04-21](/articles/india-was-the-worlds-ai-warfare-lab-heres-what-actually-happened) [// ai](/articles/qwen-just-became-the-most-downloaded-ai-model-heres-why-nobodys-talking-about-it) ### [Qwen Just Became the Most Downloaded AI Model — Here's Why Nobody's Talking About It](/articles/qwen-just-became-the-most-downloaded-ai-model-heres-why-nobodys-talking-about-it) [2026-04-17](/articles/qwen-just-became-the-most-downloaded-ai-model-heres-why-nobodys-talking-about-it) [// ai](/articles/muse-spark-metas-new-ai-model-is-good-but-not-open-source) ### [Muse Spark: Meta's New AI Model Is Good. But Not Open Source.](/articles/muse-spark-metas-new-ai-model-is-good-but-not-open-source) [2026-04-09](/articles/muse-spark-metas-new-ai-model-is-good-but-not-open-source) [// ai](/articles/claude-code-source-leak-what-390k-lines-expose-about-ais-secret-sauce) ### [Claude Code Source Leak: What 390K Lines Expose About AI's "Secret Sauce"](/articles/claude-code-source-leak-what-390k-lines-expose-about-ais-secret-sauce) [2026-04-02](/articles/claude-code-source-leak-what-390k-lines-expose-about-ais-secret-sauce) [// ai](/articles/maven-smart-system-how-silicon-valley-optimized-the-kill-chain) ### [Maven Smart System: How Silicon Valley Optimized the Kill Chain](/articles/maven-smart-system-how-silicon-valley-optimized-the-kill-chain) [2026-03-26](/articles/maven-smart-system-how-silicon-valley-optimized-the-kill-chain) [// tech](/articles/jio-hotstar-s-feature-flagging-how-they-ship-at-scale) ### [JioHotstar's Feature Flagging: How They Ship at Scale](/articles/jio-hotstar-s-feature-flagging-how-they-ship-at-scale) [2026-03-23](/articles/jio-hotstar-s-feature-flagging-how-they-ship-at-scale) [// Linux](/articles/open-claw-the-linux-of-ai-agents-or-a-security-nightmare) ### [OpenClaw: The Linux of AI Agents or a Security Nightmare?](/articles/open-claw-the-linux-of-ai-agents-or-a-security-nightmare) [2026-03-22](/articles/open-claw-the-linux-of-ai-agents-or-a-security-nightmare) [// Opinions](/articles/why-technical-depth-matters) ### [Why Technical Depth Matters More Than Content Volume](/articles/why-technical-depth-matters) [2026-03-21](/articles/why-technical-depth-matters) [// AI](/articles/mozilla-s-rebel-alliance-can-a-nonprofit-birth-a-startup-without-losing-its-soul) ### [Mozilla's Rebel Alliance: Can a Nonprofit Win the AI War?](/articles/mozilla-s-rebel-alliance-can-a-nonprofit-birth-a-startup-without-losing-its-soul) [2026-03-21](/articles/mozilla-s-rebel-alliance-can-a-nonprofit-birth-a-startup-without-losing-its-soul) [// IoT](/articles/iot-future-edge-computing) ### [The Future of IoT: Edge Computing and Beyond](/articles/iot-future-edge-computing) [2026-03-21](/articles/iot-future-edge-computing) [// Tech](/articles/apple-vision-light-the-ar-glasses-we-ve-been-waiting-for) ### [Apple Vision Light: The AR Glasses We've Been Waiting For?](/articles/apple-vision-light-the-ar-glasses-we-ve-been-waiting-for) [2026-03-20](/articles/apple-vision-light-the-ar-glasses-we-ve-been-waiting-for) [// AI](/articles/ai-models-2026) ### [Announcing Vite+ Alpha](/articles/ai-models-2026) [2026-03-13](/articles/ai-models-2026) [// Tech](/articles/web-frameworks-benchmark) ### [VoidZero and npmx: Building Better Tools Together](/articles/web-frameworks-benchmark) [2026-03-03](/articles/web-frameworks-benchmark) [// Tech](/articles/quantum-computing-breakthrough) ### [What's New in ViteLand: February 2026 Recap](/articles/quantum-computing-breakthrough) [2026-03-02](/articles/quantum-computing-breakthrough) [// Tech](/articles/tech-layoffs-stabilize) ### [Tech Industry Hiring Stabilizes After Tumultuous Year](/articles/tech-layoffs-stabilize) [2026-02-28](/articles/tech-layoffs-stabilize) [// Linux](/articles/rust-in-linux-kernel) ### [Rust in the Linux Kernel: One Year Later](/articles/rust-in-linux-kernel) [2026-01-22](/articles/rust-in-linux-kernel) No articles found. // By Category // ## Linux [View All](/articles?category=Linux) [// Linux](/articles/open-claw-the-linux-of-ai-agents-or-a-security-nightmare) ### [OpenClaw: The Linux of AI Agents or a Security Nightmare?](/articles/open-claw-the-linux-of-ai-agents-or-a-security-nightmare) [OpenClaw hit 250K GitHub stars as the "Linux of AI agents." With CVE-2026-25253, 41% vulnerable skills, and 21K exposed instances, is it infrastructure or nightmare?](/articles/open-claw-the-linux-of-ai-agents-or-a-security-nightmare) [2026-03-22](/articles/open-claw-the-linux-of-ai-agents-or-a-security-nightmare) [// Linux](/articles/rust-in-linux-kernel) ### [Rust in the Linux Kernel: One Year Later](/articles/rust-in-linux-kernel) [Reflecting on one year of Rust programming language integration into the Linux kernel development.](/articles/rust-in-linux-kernel) [2026-01-22](/articles/rust-in-linux-kernel) // ## Tech [View All](/articles?category=Tech) [// tech](/articles/jio-hotstar-s-feature-flagging-how-they-ship-at-scale) ### [JioHotstar's Feature Flagging: How They Ship at Scale](/articles/jio-hotstar-s-feature-flagging-how-they-ship-at-scale) [JioHotstar ships \~12 features a week to 500M users without breaking 61M live viewers. Here's how their feature flagging system actually works.](/articles/jio-hotstar-s-feature-flagging-how-they-ship-at-scale) [2026-03-23](/articles/jio-hotstar-s-feature-flagging-how-they-ship-at-scale) [// Tech](/articles/apple-vision-light-the-ar-glasses-we-ve-been-waiting-for) ### [Apple Vision Light: The AR Glasses We've Been Waiting For?](/articles/apple-vision-light-the-ar-glasses-we-ve-been-waiting-for) [A comprehensive review of Apple's new Vision Light AR glasses and their potential impact on the augmented reality market.](/articles/apple-vision-light-the-ar-glasses-we-ve-been-waiting-for) [2026-03-20](/articles/apple-vision-light-the-ar-glasses-we-ve-been-waiting-for) [// Tech](/articles/web-frameworks-benchmark) ### [VoidZero and npmx: Building Better Tools Together](/articles/web-frameworks-benchmark) [Exploring the collaboration between VoidZero and npmx to create better JavaScript development tools.](/articles/web-frameworks-benchmark) [2026-03-03](/articles/web-frameworks-benchmark) // ## AI [View All](/articles?category=AI) [// ai](/articles/ai-in-india-2026-scared-leading-and-wasting-water) ### [AI in India 2026: Scared, Leading, and Wasting Water](/articles/ai-in-india-2026-scared-leading-and-wasting-water) [India leads the world in enterprise AI at 80% but ranks 101st per person. The full honest picture: who's winning, what it costs, and what you should actually do.](/articles/ai-in-india-2026-scared-leading-and-wasting-water) [2026-05-16](/articles/ai-in-india-2026-scared-leading-and-wasting-water) [// ai](/articles/when-claude-became-my-relationship-coach-she-hated-ai) ### [When Claude Became My Relationship Coach (She Hated AI)](/articles/when-claude-became-my-relationship-coach-she-hated-ai) [I deleted 1,080+ messages on Claude's advice. Then she found out I was using AI to talk to her. Here's where I drew the line — and where I should've started.](/articles/when-claude-became-my-relationship-coach-she-hated-ai) [2026-05-11](/articles/when-claude-became-my-relationship-coach-she-hated-ai) [// ai](/articles/tokens-are-getting-cheaper-so-why-does-ai-feel-more-expensive) ### [Tokens Are Getting Cheaper. So Why Does AI Feel More Expensive?](/articles/tokens-are-getting-cheaper-so-why-does-ai-feel-more-expensive) [Prime's right that the AI economy is shifting — but the real story isn't about money. It's about GPUs, and nobody has enough of them. Let's break this down.](/articles/tokens-are-getting-cheaper-so-why-does-ai-feel-more-expensive) [2026-05-08](/articles/tokens-are-getting-cheaper-so-why-does-ai-feel-more-expensive) // ## IoT [View All](/articles?category=IoT) [// IoT](/articles/iot-future-edge-computing) ### [The Future of IoT: Edge Computing and Beyond](/articles/iot-future-edge-computing) [Exploring how edge computing is revolutionizing the Internet of Things and enabling real-time processing at the network's edge.](/articles/iot-future-edge-computing) [2026-03-21](/articles/iot-future-edge-computing) // ## Opinions [View All](/articles?category=Opinions) [// Opinions](/articles/why-technical-depth-matters) ### [Why Technical Depth Matters More Than Content Volume](/articles/why-technical-depth-matters) [A reflection on why in-depth technical content is more valuable than superficial quantity in tech media.](/articles/why-technical-depth-matters) [2026-03-21](/articles/why-technical-depth-matters) [Browse All Articles](/articles) --- # About Shravonix > Learn about Shravonix, a technical publication delivering in-depth analysis of Linux, AI, IoT, and software engineering for developers and tech enthusiasts. // About ## // Mission Shravonix is a technical publication dedicated to bringing you deep, thoughtful analysis of the technologies that shape our world. We believe in going beyond surface-level reporting to explore the technical foundations, implications, and future directions of emerging technologies. Our focus is on quality over quantity. Every piece we publish is researched thoroughly and written for readers who want to understand _how_ technology works, not just what's trending. ## // What We Cover 01 ### Linux Kernel updates, distributions, and open-source projects. 02 ### Tech Industry trends, product launches, and tech ecosystem analysis. 03 ### AI Artificial intelligence, machine learning, and automation. 04 ### IoT Internet of Things, embedded systems, and smart devices. 05 ### Opinions Thoughtful commentary on technology and its impact on society. ## // Team ### Jainil Prajapati Contributor [View All Authors](/authors) ## // Get in Touch Have feedback, questions, or interested in contributing? We'd love to hear from you. [Write for Us ](https://github.com/jaainil/shravonix-com/issues/new?template=GUEST_POST.md)[GitHub ](https://github.com/shravonix)[X.com ](https://x.com/shravonix_com)[Email](mailto:hello@shravonix.com) --- # Article Index > Browse every article published on Shravonix. In-depth technical analysis on AI, Linux, IoT, and software engineering. // Archive 20 articles All Linux Tech AI IoT Opinions Programming SEARCH ARTICLES... [// ai](/articles/ai-in-india-2026-scared-leading-and-wasting-water) ### [AI in India 2026: Scared, Leading, and Wasting Water](/articles/ai-in-india-2026-scared-leading-and-wasting-water) [India leads the world in enterprise AI at 80% but ranks 101st per person. The full honest picture: who's winning, what it costs, and what you should actually do.](/articles/ai-in-india-2026-scared-leading-and-wasting-water) [2026-05-16](/articles/ai-in-india-2026-scared-leading-and-wasting-water) [// ai](/articles/when-claude-became-my-relationship-coach-she-hated-ai) ### [When Claude Became My Relationship Coach (She Hated AI)](/articles/when-claude-became-my-relationship-coach-she-hated-ai) [I deleted 1,080+ messages on Claude's advice. Then she found out I was using AI to talk to her. Here's where I drew the line — and where I should've started.](/articles/when-claude-became-my-relationship-coach-she-hated-ai) [2026-05-11](/articles/when-claude-became-my-relationship-coach-she-hated-ai) [// ai](/articles/tokens-are-getting-cheaper-so-why-does-ai-feel-more-expensive) ### [Tokens Are Getting Cheaper. So Why Does AI Feel More Expensive?](/articles/tokens-are-getting-cheaper-so-why-does-ai-feel-more-expensive) [Prime's right that the AI economy is shifting — but the real story isn't about money. It's about GPUs, and nobody has enough of them. Let's break this down.](/articles/tokens-are-getting-cheaper-so-why-does-ai-feel-more-expensive) [2026-05-08](/articles/tokens-are-getting-cheaper-so-why-does-ai-feel-more-expensive) [// ai](/articles/tokens-are-cheap-thinking-isnt) ### [Tokens Are Cheap. Thinking Isn't.](/articles/tokens-are-cheap-thinking-isnt) [I put a daily limit on AI prompts. Not because I can't afford the tokens, but because my brain can't afford the attention. Here's why limiting daily AI prompts protects clarity.](/articles/tokens-are-cheap-thinking-isnt) [2026-04-24](/articles/tokens-are-cheap-thinking-isnt) [// ai](/articles/india-was-the-worlds-ai-warfare-lab-heres-what-actually-happened) ### [India Was the World's AI Warfare Lab. Here's What Actually Happened.](/articles/india-was-the-worlds-ai-warfare-lab-heres-what-actually-happened) [From the 2024 elections to Operation Sindoor, India faced the most complete AI information warfare campaign ever run against a democracy. Here's the full picture, technically. (177 chars — trim to: From deepfakes in 2024 elections to nuclear-risk AI disinfo during Sindoor — India is the world's most complete live case study in AI warfare. Here's the full technical breakdown. (181 chars — trim further:) China, the US, and AI in Indian elections — from deepfakes to Sindoor to arms market manipulation. The complete technical breakdown no one else is doing.](/articles/india-was-the-worlds-ai-warfare-lab-heres-what-actually-happened) [2026-04-21](/articles/india-was-the-worlds-ai-warfare-lab-heres-what-actually-happened) [// ai](/articles/qwen-just-became-the-most-downloaded-ai-model-heres-why-nobodys-talking-about-it) ### [Qwen Just Became the Most Downloaded AI Model — Here's Why Nobody's Talking About It](/articles/qwen-just-became-the-most-downloaded-ai-model-heres-why-nobodys-talking-about-it) [Alibaba's Qwen hit 1 billion downloads. 8 of 10 top AI models on Hugging Face are Qwen. Then the guy who built it walked out. Here's what actually happened.](/articles/qwen-just-became-the-most-downloaded-ai-model-heres-why-nobodys-talking-about-it) [2026-04-17](/articles/qwen-just-became-the-most-downloaded-ai-model-heres-why-nobodys-talking-about-it) [// ai](/articles/muse-spark-metas-new-ai-model-is-good-but-not-open-source) ### [Muse Spark: Meta's New AI Model Is Good. But Not Open Source.](/articles/muse-spark-metas-new-ai-model-is-good-but-not-open-source) [Meta Superintelligence Labs dropped Muse Spark after 9 months. Benchmarks are promising. But the open-source plot twist? Nobody's talking about it. Let's go.](/articles/muse-spark-metas-new-ai-model-is-good-but-not-open-source) [2026-04-09](/articles/muse-spark-metas-new-ai-model-is-good-but-not-open-source) [// ai](/articles/claude-code-source-leak-what-390k-lines-expose-about-ais-secret-sauce) ### [Claude Code Source Leak: What 390K Lines Expose About AI's "Secret Sauce"](/articles/claude-code-source-leak-what-390k-lines-expose-about-ais-secret-sauce) [Anthropic accidentally open-sourced their flagship AI coding tool via source maps. Here's what the leak reveals about agent architecture, internal features, and why developers are rewriting it in Rust.](/articles/claude-code-source-leak-what-390k-lines-expose-about-ais-secret-sauce) [2026-04-02](/articles/claude-code-source-leak-what-390k-lines-expose-about-ais-secret-sauce) [// ai](/articles/maven-smart-system-how-silicon-valley-optimized-the-kill-chain) ### [Maven Smart System: How Silicon Valley Optimized the Kill Chain](/articles/maven-smart-system-how-silicon-valley-optimized-the-kill-chain) [The Pentagon just made Palantir's Maven Smart System an official program of record. Here's how the AI stack behind modern warfare actually works — and why it matters.](/articles/maven-smart-system-how-silicon-valley-optimized-the-kill-chain) [2026-03-26](/articles/maven-smart-system-how-silicon-valley-optimized-the-kill-chain) [// tech](/articles/jio-hotstar-s-feature-flagging-how-they-ship-at-scale) ### [JioHotstar's Feature Flagging: How They Ship at Scale](/articles/jio-hotstar-s-feature-flagging-how-they-ship-at-scale) [JioHotstar ships \~12 features a week to 500M users without breaking 61M live viewers. Here's how their feature flagging system actually works.](/articles/jio-hotstar-s-feature-flagging-how-they-ship-at-scale) [2026-03-23](/articles/jio-hotstar-s-feature-flagging-how-they-ship-at-scale) [// Linux](/articles/open-claw-the-linux-of-ai-agents-or-a-security-nightmare) ### [OpenClaw: The Linux of AI Agents or a Security Nightmare?](/articles/open-claw-the-linux-of-ai-agents-or-a-security-nightmare) [OpenClaw hit 250K GitHub stars as the "Linux of AI agents." With CVE-2026-25253, 41% vulnerable skills, and 21K exposed instances, is it infrastructure or nightmare?](/articles/open-claw-the-linux-of-ai-agents-or-a-security-nightmare) [2026-03-22](/articles/open-claw-the-linux-of-ai-agents-or-a-security-nightmare) [// Opinions](/articles/why-technical-depth-matters) ### [Why Technical Depth Matters More Than Content Volume](/articles/why-technical-depth-matters) [A reflection on why in-depth technical content is more valuable than superficial quantity in tech media.](/articles/why-technical-depth-matters) [2026-03-21](/articles/why-technical-depth-matters) [// AI](/articles/mozilla-s-rebel-alliance-can-a-nonprofit-birth-a-startup-without-losing-its-soul) ### [Mozilla's Rebel Alliance: Can a Nonprofit Win the AI War?](/articles/mozilla-s-rebel-alliance-can-a-nonprofit-birth-a-startup-without-losing-its-soul) [Mozilla built a legal structure to stop itself from becoming the next OpenAI. Here's how the 80/20 split, PBC bylaws, and a $1.4B bet on open-source AI actually works.](/articles/mozilla-s-rebel-alliance-can-a-nonprofit-birth-a-startup-without-losing-its-soul) [2026-03-21](/articles/mozilla-s-rebel-alliance-can-a-nonprofit-birth-a-startup-without-losing-its-soul) [// IoT](/articles/iot-future-edge-computing) ### [The Future of IoT: Edge Computing and Beyond](/articles/iot-future-edge-computing) [Exploring how edge computing is revolutionizing the Internet of Things and enabling real-time processing at the network's edge.](/articles/iot-future-edge-computing) [2026-03-21](/articles/iot-future-edge-computing) [// Tech](/articles/apple-vision-light-the-ar-glasses-we-ve-been-waiting-for) ### [Apple Vision Light: The AR Glasses We've Been Waiting For?](/articles/apple-vision-light-the-ar-glasses-we-ve-been-waiting-for) [A comprehensive review of Apple's new Vision Light AR glasses and their potential impact on the augmented reality market.](/articles/apple-vision-light-the-ar-glasses-we-ve-been-waiting-for) [2026-03-20](/articles/apple-vision-light-the-ar-glasses-we-ve-been-waiting-for) [// AI](/articles/ai-models-2026) ### [Announcing Vite+ Alpha](/articles/ai-models-2026) [Announcing the alpha release of Vite+, the next generation of the Vite build tool with enhanced features and performance.](/articles/ai-models-2026) [2026-03-13](/articles/ai-models-2026) [// Tech](/articles/web-frameworks-benchmark) ### [VoidZero and npmx: Building Better Tools Together](/articles/web-frameworks-benchmark) [Exploring the collaboration between VoidZero and npmx to create better JavaScript development tools.](/articles/web-frameworks-benchmark) [2026-03-03](/articles/web-frameworks-benchmark) [// Tech](/articles/quantum-computing-breakthrough) ### [What's New in ViteLand: February 2026 Recap](/articles/quantum-computing-breakthrough) [A comprehensive recap of all the exciting updates and developments in the Vite ecosystem for February 2026.](/articles/quantum-computing-breakthrough) [2026-03-02](/articles/quantum-computing-breakthrough) [// Tech](/articles/tech-layoffs-stabilize) ### [Tech Industry Hiring Stabilizes After Tumultuous Year](/articles/tech-layoffs-stabilize) [Analysis of the tech hiring landscape in early 2026 and what it means for developers and companies.](/articles/tech-layoffs-stabilize) [2026-02-28](/articles/tech-layoffs-stabilize) [// Linux](/articles/rust-in-linux-kernel) ### [Rust in the Linux Kernel: One Year Later](/articles/rust-in-linux-kernel) [Reflecting on one year of Rust programming language integration into the Linux kernel development.](/articles/rust-in-linux-kernel) [2026-01-22](/articles/rust-in-linux-kernel) No articles found. --- # AI in India 2026: Scared, Leading, and Wasting Water > India leads the world in enterprise AI at 80% but ranks 101st per person. The full honest picture: who's winning, what it costs, and what you should actually do. // ai 2026-05-16 [Jainil Prajapati](/authors/jainil-prajapati) 2026-05-16 - [Okay, So Where Does India Actually Stand With AI?](#okay-so-where-does-india-actually-stand-with-ai) - [The Numbers That Most Headlines Are Getting Wrong](#the-numbers-that-most-headlines-are-getting-wrong) - [But Here’s the Part That Should Give You Pause](#but-heres-the-part-that-should-give-you-pause) - [Wait, I Need to Talk About the Ghibli Thing](#wait-i-need-to-talk-about-the-ghibli-thing) - [The Visakhapatnam Story Is Huge (and Not Talked About Enough)](#the-visakhapatnam-story-is-huge-and-not-talked-about-enough) - [Will AI Take Your Job? Actually Answering This](#will-ai-take-your-job-actually-answering-this) - [So Use AI Smartly, Man](#so-use-ai-smartly-man) - [FAQ](#faq) Here’s the thing about India and AI. We are simultaneously ranked #1 and #101. Not being dramatic. Anthropic published their India Country Brief earlier this year. The data is genuinely interesting. **India is the #2 country in total Claude AI usage globally.** Right behind the US. But per capita — adjusting for our working-age population — **India ranks 101st out of 116 nations.** Those two things shouldn’t both be true. Except they are. And that gap is exactly the thing worth understanding. --- ## Okay, So Where Does India Actually Stand With AI? India is the world’s #2 user of Claude AI and ranks #1 globally in AI use for software tasks at 45.2%. But per capita, India sits 101st out of 116 nations — meaning total usage is large because of population size, not because the average Indian is heavily using AI. Indian users who do use AI get a **15x productivity speedup**, compared to 12x globally. The opportunity is real but narrow. --- ## The Numbers That Most Headlines Are Getting Wrong Time for my classic data paragraph. Bear with me. The Anthropic Economic Index published a full India brief in February 2026, based on roughly a million Claude.ai conversations globally during November 2025. Some things they found: **India ranks #1 globally in software-related AI task usage.** 45.2% of all Indian Claude use maps to software occupations. Vietnam is #2. Egypt is #3. We are leading. Indian users take an average of 14.8 minutes to do tasks that would otherwise take **3.8 hours.** That’s a 15x speedup. The global average is 12x. Meaning when Indians use AI, they’re applying it to harder problems and getting more out of it. 51.3% of Indian Claude usage is work-related. Only 27.8% is personal. The global average for personal use is 34.7%. We are using AI for work more than the world average. By a meaningful margin. And enterprise AI adoption? India is at **80%**. The US is at **59%**. > India is the world’s most aggressive enterprise AI adopter — outpacing the US by 21 percentage points, at a lower GDP per capita. Yuh. Seriously. --- ## But Here’s the Part That Should Give You Pause All those numbers sound incredible. And they are. But here’s the catch. The 80% enterprise adoption? 74% of it is in BFSI, IT services, and telecom. Heavily concentrated. The total Claude usage that makes us #2 globally? Over half of it comes from just **four states: Maharashtra, Tamil Nadu, Karnataka, and Delhi.** Four states. Out of 28. Per capita, adjusted for our massive population, India sits **101st out of 116 countries** in Claude usage. That number reflects what the Anthropic team actually said: India’s high absolute usage reflects population size — not that the average Indian is heavily using AI. The people who are crushing it with AI in India are in GCCs, IT companies, top-tier universities. Everyone else is — honestly — either making Ghibli portraits or mildly scared of the whole thing. The 14% figure is worth noting here. Data shows a meaningfully higher percentage of Indians express hesitation or anxiety about AI products compared to peer countries. On the flip side, among working professionals who use AI regularly? India leads the usage-intensity charts. Split personality. Exactly. --- ## Wait, I Need to Talk About the Ghibli Thing I’m going here because nobody else is. Earlier this year, ChatGPT’s image model went viral. Everyone made Studio Ghibli-style portraits. Including me. I made multiple. I posted them on Twitter. People asked me to make theirs. I made those too. It was fun. I’m not saying it wasn’t. But every single one of those images required backend computation that you can’t see. GPU clusters running at full load. Data centers processing millions of simultaneous requests. And to run those data centers at that scale? They need **electricity.** A lot of it. And to keep those GPUs from literally overheating and melting down? They need **water.** A shocking, uncomfortable amount of water. One large Meta data center uses up to **500,000 gallons of water per day.** Drink-quality, clean water that goes in, heats up, gets vented out. Rinse and repeat, forever. The Stanford AI Index 2026 dropped this number quietly: > Annual water use for GPT-4o inference alone could exceed the drinking water needs of **12 million people.** 12 million people’s annual drinking water. For one model’s inference. And AI data center power capacity globally has hit **29.6 gigawatts** — comparable to the entire state of New York at peak demand. The Business Insider investigation on US data centers found neighborhoods in Northern Virginia where electricity bills **doubled**, drinking water depleted, constant hum vibrating people’s walls at night, families moving to basements with noise-cancelling headphones just to sleep. One resident near Amazon’s data center cluster said her 7-year-old son kept waking up from nightmares thinking there was a spaceship outside. That’s not clickbait. That’s a person’s life. Now — India is about to get a lot more of these. Which brings me to Vizag. --- ## The Visakhapatnam Story Is Huge (and Not Talked About Enough) April 28, 2026. Andhra Pradesh CM Chandrababu Naidu laid the foundation stone for **Google’s ₹1.35 lakh crore AI hub in Visakhapatnam.** $15 billion. Google’s largest ever investment in India. One of the largest single FDIs in India’s history. The specs: - Initial capacity: **1 gigawatt** - Potential scale: **5 gigawatts** - Land: 600 acres across three campuses - Partners: Adani ConneX and Airtel Nxtra - Three international subsea cables landing at Vizag — connecting India directly to the US, Europe, Africa, and Australia For reference: India’s **total national data center capacity** was approximately **1.5 GW** as of late 2025. This one project could eventually be **3x all of India’s current capacity.** Union IT Minister Ashwini Vaishnaw said it. Visakhapatnam is being “reborn as AI Patnam.” Like how Hyderabad became Cyberabad in the 90s. That’s the ambition. And honestly? I believe it. The strategic logic is real. Vizag sits on the eastern coast — proximity to Southeast Asia, Australia, the emerging digital demand centers of the world. The subsea cable gateway is genuinely significant for India’s digital sovereignty. But. A 5 GW facility needs power. It needs water. Lots of both. The environmental questions that are currently destroying neighborhoods in Virginia, draining aquifers in Arizona, doubling electricity bills in Nebraska — those questions are coming here. We need to build the infrastructure and ask the questions at the same time. Not infrastructure first, questions never. --- ## Will AI Take Your Job? Actually Answering This No more “it depends.” Let me just say the thing. Per the Anthropic labor market report: **computer programmers have a 74.5% AI exposure rate.** Customer service representatives are heavily exposed. Data entry, financial analysis, admin work — all high. Call centers? You’ve already seen it — you call Jio or Airtel and you’re talking to AI, not a person. The Stanford AI Index 2026 — published April 2026 — gives us the first hard labor data: **Entry-level software developer jobs for workers aged 22–25 fell nearly 20% since 2024.** This is the first white-collar job category with a measurable hiring contraction directly attributable to AI. And the trend is accelerating. So yes. It’s happening. But here’s the nuance that gets lost: Productivity for people who kept their jobs went up. Marketing teams using AI show productivity gains up to 72%. Software developers up 14-26%. Customer support up 14-26%. The math is simple and brutal. If what used to take 8 hours now takes 4 hours — your company doesn’t send you home early. They give you more work. Or they hire half as many people for the same output. Companies don’t exist to give people jobs. They exist to make money with as few inputs as possible. AI made one major input — human labor — cheaper and faster. Cognizant announced potential impacts to 7,000–15,000 employees in one news cycle. It’s not hypothetical anymore. My honest take: **the people who will be fine aren’t the ones avoiding AI. They’re the ones who use it so well they become irreplaceable.** Ground maintenance? Still needs humans. Courtroom lawyering? Still needs humans. Surgery? Still needs humans. But if your job is mostly information processing — writing documents, debugging code, answering queries, analyzing data — AI is already doing a version of that better than a freshly hired grad. The question is which side of that you’re on. --- ## So Use AI Smartly, Man I know this sounds simple as I’m writing it. But the data literally proves it. The longer you use AI — and the better you get at prompting, refining, iterating — the more value you extract. The Anthropic data shows high-tenure users get significantly better task success rates. The tool improves for you as you improve with it. ChatGPT and Perplexity usage in India is among the highest in the world. Indian students are near the top globally for AI tool adoption in coursework. We have the second-largest AI talent pool on the planet. The infrastructure is coming. The $67 billion that Microsoft, Amazon, and Google committed to India? It lands. The IndiaAI Mission’s 38,000 GPUs — already exceeded the target. GPU compute in India costs 40-50% less than the global average. The only variable left is you. Not building a data center. Not deploying an agentic workflow. Just — learning to use the tool properly. Getting good at it. Using it for actual work, not just for making your face look like a Studio Ghibli character. I’m not saying Ghibli was wrong. I made multiple. They were nice. I’m saying: also use it for the thing that’s going to matter in two years. --- ## FAQ **Will AI replace jobs in India?** It already has in some roles. Entry-level developer hiring fell \~20% globally since 2024. Call center and data entry roles face high exposure. Jobs requiring judgment, physical presence, or complex human relationships are safer. The bigger risk is not adapting — and watching others who do adapt become more competitive than you. **Which jobs in India are safe from AI?** Ground maintenance, healthcare requiring hands-on judgment, courtroom legal work, and physical trades are lower risk currently. Any job that’s primarily information processing, document creation, or routine customer interaction faces high exposure. Check the Anthropic labor report — field-by-field breakdown is public. **Is India really #1 in AI adoption?** Indian enterprise adoption leads the world at 80% vs the US at 59%. But per-capita individual usage ranks 101st out of 116 countries. It’s concentrated in four states and in IT services. Both things are true. We lead in intensity; we’re average in spread. **What’s actually happening in Visakhapatnam?** Google broke ground on April 28, 2026 for a $15B AI hub — India’s first gigawatt-scale data center. Initial 1 GW capacity, potential 5 GW. Three subsea cables connecting Vizag to the US, Europe, Africa, and Australia. Target commissioning: July 2028. It’s genuinely significant. **What’s the environmental cost of AI?** Significant and underreported. Data centers need massive electricity and clean water for GPU cooling. The Stanford AI Index 2026 says GPT-4o inference water use alone could exceed 12 million people’s annual drinking water needs. Global AI data center power has hit 29.6 GW. These costs are real — and they’re coming to India too. --- Anyway. The data is out there. The reports are linked. The picture is not as simple as “AI superpower India” or “AI will destroy Indian IT.” It’s weirder and more interesting than both. Use it smartly. That’s it. Chalo, bye. **PS:** The next time you’re batch-generating Ghibli portraits for your entire contact list — maybe just make one. The GPU is tired. Kinda. ## Comments ## Related Articles [// ai](/articles/when-claude-became-my-relationship-coach-she-hated-ai) ### [When Claude Became My Relationship Coach (She Hated AI)](/articles/when-claude-became-my-relationship-coach-she-hated-ai) [2026-05-11](/articles/when-claude-became-my-relationship-coach-she-hated-ai) [// ai](/articles/tokens-are-getting-cheaper-so-why-does-ai-feel-more-expensive) ### [Tokens Are Getting Cheaper. So Why Does AI Feel More Expensive?](/articles/tokens-are-getting-cheaper-so-why-does-ai-feel-more-expensive) [2026-05-08](/articles/tokens-are-getting-cheaper-so-why-does-ai-feel-more-expensive) [// ai](/articles/tokens-are-cheap-thinking-isnt) ### [Tokens Are Cheap. Thinking Isn't.](/articles/tokens-are-cheap-thinking-isnt) [2026-04-24](/articles/tokens-are-cheap-thinking-isnt) ### Never miss an update Join 50,000+ developers getting our weekly tech insights. [YOUR EMAIL ADDRESS]() Subscribe --- # Announcing Vite+ Alpha > Announcing the alpha release of Vite+, the next generation of the Vite build tool with enhanced features and performance. // AI 2026-03-13 [Jainil Prajapati](/authors/jainil-prajapati) 2026-03-13 - [Introduction](#introduction) - [Key Takeaways](#key-takeaways) - [Deep Dive](#deep-dive) - [Future Outlook](#future-outlook) - [Conclusion](#conclusion) The tech landscape is evolving faster than ever. In this comprehensive overview, we break down the most significant developments and what they mean for the future of the industry. As we move further into 2026, the convergence of artificial intelligence, advanced hardware, and new web paradigms is creating unprecedented opportunities and challenges. Companies are racing to adapt to a reality where AI is no longer just a feature, but the foundation of new products. ## Introduction This shift requires a fundamental rethinking of how we build, deploy, and scale applications. The tooling ecosystem has matured significantly, with Vite emerging as the de facto standard for modern web development. ### Key Takeaways Our analysis of recent industry trends reveals several critical insights that developers and tech leaders need to understand: - **AI-First Architecture:** Designing systems with machine learning models at their core - **Edge Computing:** Pushing processing closer to the user for lower latency - **Security by Default:** Implementing zero-trust frameworks across all layers > “The companies that succeed in the next decade won’t just use AI; they will be fundamentally restructured around it.” ## Deep Dive Let’s look at the numbers. Recent benchmarks show a massive improvement in processing efficiency for large language models, reducing the cost of inference by over 60% compared to last year. This democratization of AI capabilities means smaller teams can now build features that previously required massive engineering organizations. ```javascript // Example of a modern API integration async function fetchInsights() { const response = await fetch('https://api.techwire.dev/v1/insights', { headers: { 'Authorization': `Bearer ${process.env.API_KEY}`, 'Content-Type': 'application/json' } }); return response.json(); } ``` ## Future Outlook Looking ahead, we expect to see continued consolidation in the tooling space. Developers are tired of configuring complex build pipelines and are migrating towards unified, zero-config toolchains that “just work.” The Vite+ project represents our vision for the next generation of build tooling—faster, simpler, and more powerful than ever before. ## Conclusion The pace of innovation isn’t slowing down. To stay competitive, teams must remain agile, continuously evaluate their tech stacks, and be willing to adopt new paradigms when they offer clear advantages. ## Comments ## Related Articles [// Tech](/articles/quantum-computing-breakthrough) ### [What's New in ViteLand: February 2026 Recap](/articles/quantum-computing-breakthrough) [2026-03-02](/articles/quantum-computing-breakthrough) [// Tech](/articles/web-frameworks-benchmark) ### [VoidZero and npmx: Building Better Tools Together](/articles/web-frameworks-benchmark) [2026-03-03](/articles/web-frameworks-benchmark) [// ai](/articles/ai-in-india-2026-scared-leading-and-wasting-water) ### [AI in India 2026: Scared, Leading, and Wasting Water](/articles/ai-in-india-2026-scared-leading-and-wasting-water) [2026-05-16](/articles/ai-in-india-2026-scared-leading-and-wasting-water) ### Never miss an update Join 50,000+ developers getting our weekly tech insights. [YOUR EMAIL ADDRESS]() Subscribe --- # Apple Vision Light: The AR Glasses We've Been Waiting For? > A comprehensive review of Apple's new Vision Light AR glasses and their potential impact on the augmented reality market. // Tech 2026-03-20 [Jainil Prajapati](/authors/jainil-prajapati) 2026-03-20 - [Introduction](#introduction) - [Design and Build Quality](#design-and-build-quality) - [Key Features](#key-features) - [Performance](#performance) - [Conclusion](#conclusion) The tech landscape is evolving faster than ever. In this comprehensive review, we break down Apple’s latest venture into augmented reality and what it means for consumers and developers alike. !\[Apple Vision Light]\() ## Introduction Apple has finally unveiled its much-anticipated AR glasses, promising a seamless blend of digital and physical worlds. After years of speculation, the Vision Light represents Apple’s vision for the future of personal computing. ## Design and Build Quality True to Apple’s form, the Vision Light features premium materials and an elegant design that wouldn’t look out of place in a fashion boutique. The glasses weigh just 35 grams, making them comfortable for extended wear. ### Key Features - **Weight:** 35 grams - **Battery Life:** 8 hours continuous use - **Display:** 4K micro-OLED per eye - **Field of View:** 120 degrees - **Spatial Audio:** Built-in speakers ## Performance The custom Apple silicon powering the Vision Light delivers smooth, low-latency rendering that makes digital content feel truly integrated with the real world. ## Conclusion Apple Vision Light represents a significant step forward for consumer AR technology. While the price point remains premium, the combination of design, performance, and ecosystem integration makes it the most compelling AR glasses yet. ## Comments ## Related Articles [// ai](/articles/ai-in-india-2026-scared-leading-and-wasting-water) ### [AI in India 2026: Scared, Leading, and Wasting Water](/articles/ai-in-india-2026-scared-leading-and-wasting-water) [2026-05-16](/articles/ai-in-india-2026-scared-leading-and-wasting-water) [// ai](/articles/when-claude-became-my-relationship-coach-she-hated-ai) ### [When Claude Became My Relationship Coach (She Hated AI)](/articles/when-claude-became-my-relationship-coach-she-hated-ai) [2026-05-11](/articles/when-claude-became-my-relationship-coach-she-hated-ai) [// ai](/articles/tokens-are-getting-cheaper-so-why-does-ai-feel-more-expensive) ### [Tokens Are Getting Cheaper. So Why Does AI Feel More Expensive?](/articles/tokens-are-getting-cheaper-so-why-does-ai-feel-more-expensive) [2026-05-08](/articles/tokens-are-getting-cheaper-so-why-does-ai-feel-more-expensive) ### Never miss an update Join 50,000+ developers getting our weekly tech insights. [YOUR EMAIL ADDRESS]() Subscribe --- # Claude Code Source Leak: What 390K Lines Expose About AI's "Secret Sauce" > Anthropic accidentally open-sourced their flagship AI coding tool via source maps. Here's what the leak reveals about agent architecture, internal features, and why developers are rewriting it in Rust. // ai 2026-04-02 [Jainil Prajapati](/authors/jainil-prajapati) 2026-04-02 - [What Actually Leaked (And How)](#what-actually-leaked-and-how) - [The Features That Weren’t Supposed to Be Public](#the-features-that-werent-supposed-to-be-public) - [The Security Culture That Enabled This](#the-security-culture-that-enabled-this) - [The Community Response: Rewrite, Don’t Fork](#the-community-response-rewrite-dont-fork) - [Why This Matters: The Harness Is the Product](#why-this-matters-the-harness-is-the-product) - [What Happens Next](#what-happens-next) - [Practical Takeaways](#practical-takeaways) - [FAQ](#faq) - [The Real Story](#the-real-story) Anthropic has a leak problem. Last week, internal documents spilled details about their unreleased “Mythos” model. This week, they published the entire source code for Claude Code—their flagship AI coding agent—directly to npm. Not a hacker breach. Not a disgruntled employee. Just a source map file that wasn’t supposed to ship. Here’s the thing: this isn’t the first time. Claude Code’s source has leaked before, and Anthropic’s lawyers sent hundreds of DMCA takedowns to GitHub repos that mirrored it. But this time? The leak is bigger, the community is faster, and the narrative has shifted from “oops” to “open source by accident.” ## What Actually Leaked (And How) On March 30, 2026, Anthropic published Claude Code v2.1.89 to npm. Bundled inside was `cli.js.map`—a 60MB source map file containing the complete, unminified TypeScript source. If you downloaded the package before they yanked it, you got 390,000 lines of production code, internal comments, and unreleased features. **Source maps are debugging bridges.** When you ship minified JavaScript, you include a map file that translates compressed code back to the original source for error tracking. Normally, these maps get uploaded to Sentry or similar services—not bundled with the public package. But someone at Anthropic configured their build pipeline wrong, and the map went out with the release. The file includes everything: the main React-based terminal renderer, 40+ tools, sub-agent orchestration, and a background engine called Dream. It also contains feature flags, internal codenames (like “Tangu” for analytics), and a subsystem called “undercover mode”—ironically designed to prevent Anthropic employees from leaking internal info in public commits. ## The Features That Weren’t Supposed to Be Public Most coverage stops at “source code leaked.” But the interesting part is what the code reveals about where Anthropic is heading. **Kairos Mode**: Not just a coding assistant, but an always-on agent that “watches, logs, and proactively acts.” It maintains append-only daily logs, runs on a heartbeat timer, and can trigger actions without user input. The prompt explicitly frames this as “Claude trying to be helpful without being annoying.” It’s designed for brief interactions, scheduled check-ins, and persistent sessions—basically Jarvis for your terminal. **Dream System**: A background memory consolidation engine that runs as a sub-agent. It activates when: (1) 24 hours have passed since the last dream, (2) at least five sessions have occurred, and (3) no other dream is running. The prompt tells Claude to “synthesize what you have learned recently into durable, well-organized memory.” This is how they plan to make long-term context actually work without blowing up token costs. **Coordinator Mode**: Spins up multiple worker agents in parallel, each with full tool access but specific instructions. Think of it as Claude managing a team of Claudes. The code shows five levels of permission cascading (policy → flags → local → project → user), suggesting they’re building enterprise-grade access controls. **Buddy System**: A Tamagotchi-style companion that hatches in your terminal. It’s a deterministic gacha system with species, rarity, shiny variants, and procedurally generated stats (debugging, patience, chaos, wisdom, snark). The leak revealed it’s tied to your userId and a fixed salt—meaning it’s trivially brute-forceable. The community has already generated “god-roll” UUIDs for legendary shinies. ## The Security Culture That Enabled This This leak didn’t happen in a vacuum. Check Point Research disclosed three critical vulnerabilities in Claude Code just last month (CVE-2025-59536, CVE-2026-21852) that allowed remote code execution and API key theft via malicious `.claude/settings.json` files. Anthropic patched them, but the pattern is revealing: configuration files are treated as executable code, and the trust model assumes developers only open trusted repos. The same lax release hygiene that let those vulnerabilities ship is what put source maps in a production npm package. When your entire company is moving at breakneck speed to ship agentic features, basic packaging checks get skipped. The “undercover mode” system—designed to prevent leaks—actually confirms how paranoid Anthropic is about exposure, yet they still shipped the digital equivalent of leaving the keys in the door. ## The Community Response: Rewrite, Don’t Fork Anthropic’s legal team immediately started firing DMCA notices at GitHub repos that mirrored the leaked source. But here’s where it gets clever: developers aren’t forking the code—they’re rewriting it. One project has already translated Claude Code into Python, and another is building a Rust version. Because it’s a derivative work of leaked code, it exists in a copyright gray area. Anthropic can’t easily DMCA it, and the community gets a clean-room implementation they can actually use. This is the same strategy that let Clean Room BIOS clones flourish in the 80s, and it’s happening in real-time on GitHub. The Discord servers are buzzing with people unlocking features. Kairos mode is being activated. The Buddy system is getting modded. Someone even got Doom running inside Claude Code’s terminal renderer. It’s not just a leak—it’s a permission slip to hack. ## Why This Matters: The Harness Is the Product Here’s what nobody’s saying clearly: **The model is not the product. The harness is.** Claude Opus is the engine. Claude Code is the car. And Anthropic just gave everyone the factory schematics. The code shows exactly how they handle prompt caching, sub-agent orchestration, tool calling, memory compaction, and permission systems. For anyone building AI agents, this is a masterclass in production-ready architecture. The irony? Claude Code isn’t even the best harness. Terminal.bench ranks it 39th among harness-model pairs. Cursor’s harness gets 93% performance out of Opus vs. Claude Code’s 77%. OpenCode (open source) is arguably better architected. But Claude Code is the _most popular_, and now everyone can see how the sausage is made. ## What Happens Next Anthropic is in a corner. They can: 1. **Double down on DMCAs** and become the “bad guy” lab that sues its own users. 1. **Open source it** and lean into the momentum (but risk losing control). 1. **Ignore it** and hope the news cycle moves on (it won’t). Their official statement called it “human error, not a security breach” and promised “measures to prevent recurrence.” But the code is already mirrored on dozens of sites. You can’t un-leak 390,000 lines. The smarter move? Do what OpenAI did when their front-end code had a bug—make a joke about it. Let engineers blog about the cool features. Open source it on their own terms. The community wants to be excited about Anthropic; they’re just waiting for permission. ## Practical Takeaways **If you’re a developer:** - You can inspect the code for educational purposes. Don’t deploy it commercially—Anthropic’s terms still apply. - Look at how they structure sub-agents and prompt caching. That’s the gold. - The Buddy system is a fun Easter egg. Generate your god-roll UUID and enjoy your legendary owl companion. **If you’re building an AI tool:** - Study the permission cascade system. It’s over-engineered but solves real enterprise problems. - The Dream memory system is a clever hack for long-term context. Adapt it. - Don’t rely on obfuscation. If it’s in the client, assume it’s public. **If you’re Anthropic:** - Stop sending DMCAs to people who aren’t distributing the original code. It’s making you look scared. - Open source Claude Code. The secret is out, and the community is doing it for you anyway. - Let your engineers talk. The code is good (7/10, per Claude’s own assessment). Let them be proud. ## FAQ **Q: Is Claude Code now open source?** A: No. The code was accidentally published, but Anthropic hasn’t changed the license. Using it commercially violates their terms. However, derivative rewrites (Python/Rust ports) exist in a legal gray area. **Q: What is Kairos mode?** A: An unreleased always-on assistant mode that runs in the background, logs activity, and can act proactively. It uses scheduled check-ins and a simplified UI for non-coding tasks. **Q: How do I check if my version has the source map?** A: Download the npm package and look for `cli.js.map`. If it’s there and larger than 50MB, it’s the leaked version. Anthropic has pulled the bad release, but mirrors exist. **Q: Can Anthropic sue me for looking at the code?** A: Probably not for just looking. But redistributing the original TypeScript source or using it to build a competing product puts you in legal jeopardy. The rewrites are safer but not risk-free. **Q: What’s the difference between this and the Mythos leak?** A: The Mythos leak was internal documents about a new model. This is the actual source code for Claude Code, their developer tool. Both stem from release process failures. **Q: Is Claude Code secure to use?** A: The disclosed vulnerabilities are patched, but the leak reveals a culture of moving fast and breaking things. If you’re doing sensitive work, audit your `.claude/settings.json` and don’t open untrusted repositories. ## The Real Story This isn’t about a mistake. It’s about a fundamental shift in how AI tools are built and distributed. The harness—the glue that turns a language model into an agent—is becoming infrastructure. Infrastructure wants to be open. Anthropic tried to keep it closed, and the internet routed around them. The code is out there. The features are unlocked. The community is building. The question isn’t whether Claude Code will be open source—it’s whether Anthropic gets to be part of that conversation, or watches from the outside while their own tool gets rebuilt without them. They can still own the best model. They can still run the best API. But the harness? That belongs to everyone now. ## Comments ## Related Articles [// ai](/articles/ai-in-india-2026-scared-leading-and-wasting-water) ### [AI in India 2026: Scared, Leading, and Wasting Water](/articles/ai-in-india-2026-scared-leading-and-wasting-water) [2026-05-16](/articles/ai-in-india-2026-scared-leading-and-wasting-water) [// ai](/articles/when-claude-became-my-relationship-coach-she-hated-ai) ### [When Claude Became My Relationship Coach (She Hated AI)](/articles/when-claude-became-my-relationship-coach-she-hated-ai) [2026-05-11](/articles/when-claude-became-my-relationship-coach-she-hated-ai) [// ai](/articles/tokens-are-getting-cheaper-so-why-does-ai-feel-more-expensive) ### [Tokens Are Getting Cheaper. So Why Does AI Feel More Expensive?](/articles/tokens-are-getting-cheaper-so-why-does-ai-feel-more-expensive) [2026-05-08](/articles/tokens-are-getting-cheaper-so-why-does-ai-feel-more-expensive) ### Never miss an update Join 50,000+ developers getting our weekly tech insights. [YOUR EMAIL ADDRESS]() Subscribe --- # India Was the World's AI Warfare Lab. Here's What Actually Happened. > From the 2024 elections to Operation Sindoor, India faced the most complete AI information warfare campaign ever run against a democracy. Here's the full picture, technically. (177 chars — trim to: From deepfakes in 2024 elections to nuclear-risk AI disinfo during Sindoor — India is the world's most complete live case study in AI warfare. Here's the full technical breakdown. (181 chars — trim further:) China, the US, and AI in Indian elections — from deepfakes to Sindoor to arms market manipulation. The complete technical breakdown no one else is doing. // ai 2026-04-21 [Jainil Prajapati](/authors/jainil-prajapati) 2026-04-21 - [So what actually happened? The 40-word version.](#so-what-actually-happened-the-40-word-version) - [Why India? Because it’s worth destabilizing.](#why-india-because-its-worth-destabilizing) - [China’s Playbook: Doctrine, Not Improvisation](#chinas-playbook-doctrine-not-improvisation) - [The AI News Anchors Nobody Talked About](#the-ai-news-anchors-nobody-talked-about) - [The Manipur Vector: Seeding Separatism with AI](#the-manipur-vector-seeding-separatism-with-ai) - [China’s Typhoon Problem: The Cyber-Physical Layer](#chinas-typhoon-problem-the-cyber-physical-layer) - [DeepSeek: The Embedded Threat Nobody Is Taking Seriously Enough](#deepseek-the-embedded-threat-nobody-is-taking-seriously-enough) - [The 2024 Lok Sabha Elections: The World’s First Generative AI Electoral Event](#the-2024-lok-sabha-elections-the-worlds-first-generative-ai-electoral-event) - [The US Angle: A Different Architecture Entirely](#the-us-angle-a-different-architecture-entirely) - [The RLHF Problem: How Bias Gets Baked into the Models](#the-rlhf-problem-how-bias-gets-baked-into-the-models) - [The Opposition Co-option Mechanism: The Part Everyone Ignores](#the-opposition-co-option-mechanism-the-part-everyone-ignores) - [Operation Sindoor: The First AI War in a Nuclear Context](#operation-sindoor-the-first-ai-war-in-a-nuclear-context) - [Pakistan’s ISPR Machine: The Institutionalised Disinformation Apparatus](#pakistans-ispr-machine-the-institutionalised-disinformation-apparatus) - [They Deepfaked India’s Entire Leadership. All of Them.](#they-deepfaked-indias-entire-leadership-all-of-them) - [The General Malik Moment: ISI Caught on Record](#the-general-malik-moment-isi-caught-on-record) - [They Also Deepfaked Trump](#they-also-deepfaked-trump) - [Pakistan’s Fake Naval War: Frigates vs. Fiction](#pakistans-fake-naval-war-frigates-vs-fiction) - [The “Recycled Reality” Playbook in Full Effect](#the-recycled-reality-playbook-in-full-effect) - [The Textbook Rewrite: History Revised in Real-Time](#the-textbook-rewrite-history-revised-in-real-time) - [The Coordinated Account Networks: How the Infrastructure Actually Worked](#the-coordinated-account-networks-how-the-infrastructure-actually-worked) - [The Turkey-China-Pakistan Cognitive Triangle: One Interoperable Pressure System](#the-turkey-china-pakistan-cognitive-triangle-one-interoperable-pressure-system) - [India’s Counter-Punch: Blocking the State Media Nodes](#indias-counter-punch-blocking-the-state-media-nodes) - [The Cyber Dimension: 1.5 Million Attacks and the War India Mostly Won Quietly](#the-cyber-dimension-15-million-attacks-and-the-war-india-mostly-won-quietly) - [Operation CyberShakti: India’s Counter-Punch](#operation-cybershakti-indias-counter-punch) - [The Bangladesh and Multi-Front Coalition Angle](#the-bangladesh-and-multi-front-coalition-angle) - [China’s Arms Market Play: The Most Underreported Angle](#chinas-arms-market-play-the-most-underreported-angle) - [The Nuclear Escalation Warning](#the-nuclear-escalation-warning) - [What India Is Building: The Sovereign AI Response](#what-india-is-building-the-sovereign-ai-response) - [IndiaAI Mission 2.0: The Numbers That Matter](#indiaai-mission-20-the-numbers-that-matter) - [What Every Other Democracy Needs to Build](#what-every-other-democracy-needs-to-build) - [FAQ](#faq) During Operation Sindoor in May 2025, a video circulated showing Pakistani Prime Minister Shehbaz Sharif conceding defeat on camera. Lamenting that China and the UAE had abandoned him. Looking defeated. The problem? It was a deepfake. The original video showed Sharif _commending_ the Pakistan Air Force. AI voice cloning and lip-sync tech had replaced everything he actually said. That’s the story in miniature. India — the world’s largest democracy — has been the most intensively targeted nation in the history of **AI-driven information warfare**. Not theoretically. Not as a case study in a research paper. Actually, operationally, right now. And the playbook used against it is the one every other democracy is about to face. --- ## So what actually happened? The 40-word version. China deployed autonomous AI botnets, deepfake news anchors, and multilingual LLM-translated propaganda to destabilize India’s 2024 elections and exploit the Manipur crisis. The US — primarily under the Biden administration — operated through algorithmic bias baked into foundational AI models and NGO funding pipelines that shaped electoral narratives. During Operation Sindoor, both combined to run the world’s first AI information war in a nuclear context. Pakistan’s ISI ran the tactical layer. China ran the strategic amplification. And the domestic opposition on both sides got used as unwitting distribution infrastructure. --- ## Why India? Because it’s worth destabilizing. Here’s the thing. India is a $4 trillion economy, the Global South’s presumptive diplomatic leader, a Quad member tightening ties with Washington, and a nuclear state sharing borders with both China and Pakistan. For Beijing, a stable pro-Western India is a strategic nightmare. For anyone wanting regional dominance, India’s internal fault lines — caste, religion, linguistic diversity across 22 official languages, thousands of regional dialects — are **attack surfaces**. And generative AI just made exploiting those fault lines essentially free. --- ## China’s Playbook: Doctrine, Not Improvisation Let’s be specific about this because it gets lost in generic “foreign interference” discourse. China’s approach to information warfare is formalized doctrine. The 2008 Chinese Defence White Paper explicitly frames **“informatized warfare”** as a primary battlespace. Under Xi, the People’s Liberation Army and the United Front Work Department have formally integrated AI-generated narrative management as a pre-conflict tool. The operational engine is **Storm-1376** — also called Spamouflage and Dragonbridge. This network has been running since 2017. It spans 175+ websites across 58 languages. As of early 2026, it remains the most prolific pro-CCP influence operation ever documented. And it evolved from clunky troll farms to something genuinely frightening. > **The upgrade that matters:** Storm-1376 now uses LLMs for automated translation and contextual expansion. It scrapes Chinese-language content, runs it through large language models, and seeds culturally-nuanced, grammatically-clean posts in Hindi, Meitei, Bengali, Kannada — simultaneously — before domestic fact-checkers have woken up. That’s not a bot farm. That’s a **cognitive supply chain**. The Microsoft Threat Analysis Center (MTAC) formally warned in April 2024 — weeks before India’s Lok Sabha election — that China would “at a minimum, create and amplify AI-generated content to benefit its interests” during Indian elections. Clint Watts, MTAC’s General Manager, put it plainly: China’s AI influence operations “may prove more effective down the line.” He was right. --- ### The AI News Anchors Nobody Talked About Storm-1376 used commercial AI video platforms — specifically Synthesia — to create hyper-realistic deepfake news anchors operating under fictitious outlet “Wolf News.” These synthetic anchors delivered polished, multilingual propaganda designed to look like legitimate international journalism. In one documented operation, these AI anchors alleged that “the United States and India were responsible for the unrest in Myanmar.” A clean, CCP-aligned narrative, delivered through what looked like independent broadcast journalism. This is “narrative laundering” at industrial scale. And the detection problem is brutal: the content doesn’t look foreign. It doesn’t look bot-generated. It looks like a news segment. --- ### The Manipur Vector: Seeding Separatism with AI When ethnic violence erupted between Meitei and Kuki communities in Manipur in May 2023, China’s digital apparatus activated within days. CCP-controlled accounts — using AI-generated profile images of Western personas — began propagating the “Little China” narrative: that Manipur was historically separate from India, culturally aligned with China, and flying a “six-star red flag.” Claims of concentration camps run by the Indian state followed. The technical sophistication was the translation pipeline. Narratives originating on Douyin were NLP-translated into English, Hindi, and regional dialects and seeded onto X and YouTube — reaching the specific Meitei communities engaged in the conflict, who don’t speak Chinese. The strategic objective: foster insurgencies on India’s northeastern borders. Tie down Indian military and administrative resources in protracted internal security operations. And do it while publicly claiming to seek “regional stability.” --- ### China’s Typhoon Problem: The Cyber-Physical Layer Beyond influence operations, China runs persistent cyber intrusions under the “Typhoon” umbrella. These aren’t passive espionage. They’re pre-positioned for disruption. - **Flax Typhoon:** Targeted Indian telecom infrastructure alongside Philippines and Hong Kong operations in late 2023. - **Volt Typhoon:** Breached Indian internet firms via a Versa Networks zero-day vulnerability in 2024. - **Salt Typhoon:** Targeted Indian telecom via Cisco devices alongside 100+ countries. The Washington Post investigation revealed Chinese hackers accessed **95.2 gigabytes of immigration data** from the Indian government, with leaked files posted to GitHub. And in late 2025, Anthropic revealed that Chinese state-linked hackers had weaponized the Claude AI system to automate a global espionage campaign targeting \~30 organizations across tech, finance, chemicals, and government agencies. The shift: AI isn’t just generating content. It’s becoming an **autonomous offensive agent** in intelligence operations. --- ### DeepSeek: The Embedded Threat Nobody Is Taking Seriously Enough In January 2025, China launched DeepSeek — a low-cost, open-source AI model with near-zero barrier to access. India’s CERT-In launched an investigation after reports emerged that DeepSeek collects behavioral tracking data through prompts, battery usage, app activity, and keystrokes. Italy, Australia, and multiple US federal agencies banned it on official devices. India’s Finance Ministry issued internal advisories. A CERT-In official indicated a ban was imminent. Let me be real about what this means: if DeepSeek had significant uptake among Indian civil servants, campaign operatives, or party workers before those bans landed, that’s not a data privacy concern. That’s intelligence infrastructure. --- ## The 2024 Lok Sabha Elections: The World’s First Generative AI Electoral Event 968 million registered voters. Seven phases. $16 billion in campaign spend. And — conservatively — **$50 million funneled specifically into AI-generated political content** by domestic parties alone. What that looked like on the ground: Voice clones of deceased politicians endorsing live candidates. AI-resurrected leaders speaking in regional dialects they never spoke. Audio deepfakes distributed via \~5,800 WhatsApp groups, reaching **over 15 million people** — bypassing every platform moderation system because they lived in encrypted, dark-social environments. Audio deepfakes were considered _more_ convincing than video counterparts by the electorate. Because a voice sounds more intimate. More private. More real. The foreign contribution to this chaos? China deployed LLM translation to inject narratives about EVM fraud, anti-government rhetoric, and religious tension simultaneously across linguistic demographics — before any domestic actor could even identify the source. --- ## The US Angle: A Different Architecture Entirely China’s interference is covert, adversarial, and attributable through cyber forensics. The US interference pattern is something else. In February 2025, Donald Trump publicly revealed — then cancelled — a **$21 million USAID grant** that the Biden administration had allocated, which Trump alleged was designed to influence voter turnout in India during the 2024 Lok Sabha elections. > _“Why do we need to spend $21 million on voter turnout in India? I guess they were trying to get somebody else elected.”_ — President Donald Trump India’s MEA responded with unusually blunt language: “These are obviously very deeply troubling.” They confirmed agencies were investigating. The full picture is messier. The Indian Express found the disputed USAID funds were actually allocated to Bangladesh. But the DisInfo Lab report — _“The Invisible Hands: Foreign Interference in Indian Elections 2024”_ — alleged that entities including the Henry Luce Foundation and George Soros’ Open Society Foundation funded academic and media projects that shaped electoral narratives in India. Yuh. That’s not a bot army. That’s a narrative pipeline with institutional credibility. --- ### The RLHF Problem: How Bias Gets Baked into the Models This is the part that doesn’t get covered. And it’s arguably the most structurally dangerous vector. Large language models — ChatGPT, Gemini, the AI systems billions of people use as information oracles — are aligned via **Reinforcement Learning from Human Feedback (RLHF)**. The “reward model” is trained on data that defines the “sensed will” of humanity. That training data comes heavily from Western media corpora, academic literature, and institutional reports. Those institutional reports are produced by think tanks — many of which receive funding from US government entities like USAID and the National Science Foundation. The V-Dem Institute, which received USAID and NSF funding, categorized India as an **“electoral autocracy”** in a widely-cited report. That classification gets ingested into LLM training data. The model’s latent space becomes mathematically biased against India’s democratic institutions. The result showed up explicitly in early 2024: when users asked Google’s Gemini whether the Indian Prime Minister was a “fascist,” it **said yes** — generating a comprehensive list of accusations. When identical queries were posed about Donald Trump or Xi Jinping, the model refused to engage. Same question. Asymmetric output. This is not a Google engineer deliberately interfering in Indian elections. It’s the **automated mathematical consequence** of biased training data — laundered through an institutional pipeline and embedded into the foundational AI systems hundreds of millions of people treat as truth. Bruh. --- ## The Opposition Co-option Mechanism: The Part Everyone Ignores Okay so here’s where it gets genuinely uncomfortable. Foreign actors — especially China — don’t need to build new channels to spread their narratives. They have a more efficient approach: **find existing domestic political conflicts and amplify them**. The mechanism works like this: An opposition party is doing what opposition parties do — criticizing the incumbent, amplifying governance failures, running attack ads. Normal democracy stuff. But somewhere in that stack of content, there’s a deepfake that originated from a Spamouflage botnet. Or a “leaked document” that was actually exfiltrated and altered by APT36’s TAG-140 affiliate. Or a narrative that emerged from a foreign-funded academic paper that found its way into party communications. The opposition shares it. Millions see it. The narrative gains domestic legitimacy. And now China’s geopolitical objective has been laundered through what looks like indigenous democratic opposition. **This does not mean all opposition activity is foreign-controlled.** That’s not the point. The point is that domestic political actors — both ruling parties and opposition — need **active threat protocols** to audit whether the content they’re amplifying has foreign fingerprints on it. Because right now, the incentive to share damaging content about rivals is so strong that most parties don’t stop to ask where it came from. > When a party amplifies a foreign-seeded deepfake to score domestic political points, it is actively facilitating the subversion of its own nation’s sovereignty. Not intentionally. But the effect is identical. The reverse also applies: when the incumbent accuses opposition of foreign treason based on internationally-generated criticism — some of which may itself have been seeded to _provoke_ that exact response — foreign AI botnets benefit from the paralysis that follows. Both sides fighting. Both sides amplifying. Foreign actors collecting. That’s the design. --- ## Operation Sindoor: The First AI War in a Nuclear Context May 2025. The Pahalgam terrorist attack kills 26 Indian civilians. India launches Operation Sindoor — precision strikes deep into Pakistani territory. Pakistan retaliates under Operation Bunyan-Um-Marsoos. The largest Beyond Visual Range air combat engagement in modern military history. 114+ aircraft. French Rafales with Meteor missiles against Chinese J-10Cs with PL-15E missiles. And while all that was happening in the physical domain, a completely parallel war was being fought in the cognitive domain. --- ### Pakistan’s ISPR Machine: The Institutionalised Disinformation Apparatus Here’s the piece that most coverage skips entirely. Pakistan’s information warfare capability isn’t improvised. Since the Balakot airstrikes in 2019, the ISI and **Inter-Services Public Relations (ISPR)** have built a **dedicated ecosystem** for narrative manipulation — coordinating Pakistani media houses, digital troll networks, and social media amplification infrastructure to target domestic and international audiences simultaneously. The three-step playbook used during Operation Sindoor was explicit and documented: 1. State-linked Pakistani actors spread the narrative that Pahalgam was a **false flag operation** by India against its own citizens 1. Bot networks, AI-generated memes, and state media flooded platforms to validate the claim 1. Pakistan’s official political leadership personally amplified unverified claims — Information Minister Attaullah Tarar publicly endorsed a fabricated claim of the Indian Army surrendering at Chora Post **without any evidence whatsoever** A cabinet minister. Personally. Amplifying a fake. During Operation Sindoor, a large Pakistani spy ring was also busted that included YouTubers and social media workers operating directly for the ISI — demonstrating how actively the Pakistani establishment had integrated civilian content creators into its information warfare pipeline. Indian security agencies additionally alleged the ISI specifically designed campaigns to flood social media with false Islamophobia narratives targeting PM Modi — aimed at Gulf countries, attempting to drive a wedge between India and its Gulf allies. --- ### They Deepfaked India’s Entire Leadership. All of Them. Not one or two targets. The whole chain of command. Deepfake videos of **PM Modi, External Affairs Minister Jaishankar, and Home Minister Amit Shah** — all showing them apologizing to Pakistan, all showing them admitting India “lost the battle” — went viral across Pakistani social media simultaneously. The Urdu captions roughly translated to: _“Modi’s screams have erupted, should we accept his apology?”_ BOOM ran all three through deepfake detection tools. Results: unnatural visuals, cloned voices, high AI manipulation probability confirmed across every single video. PIB officially classified them as a Pakistani attempt to spread panic and demoralize the Indian public. Not gonna lie — the coordination here is the thing that should concern you. This wasn’t a lone operator. You don’t fabricate AI deepfakes of three senior political figures simultaneously, with synchronized social media distribution and Urdu-language captioning optimized for virality, unless someone planned it in advance. --- ### The General Malik Moment: ISI Caught on Record This one is worth slowing down for. Pakistan circulated a deepfake of former Indian Army Chief **General Ved Prakash Malik** — making him appear to say: _“Pakistan has better weapons and equipment, destruction of Rafales and S-400 is a testament to their superiority.”_ The original ANI video showed General Malik saying **the exact opposite** — that India has better weapons and equipment than Pakistan. They inverted a real statement, cloned his voice, and released it. And then General Malik himself publicly responded. > _“It is a deepfake. Pakistani ISI at work!”_ That’s a former Chief of Army Staff, on record, naming the ISI by name, in response to a synthetic video of his own face saying things he never said. I genuinely don’t know what more direct attribution looks like. That quote should be in every article written about AI-enabled state disinformation. Every framework document. Every policy brief. Full stop. --- ### They Also Deepfaked Trump Because apparently the Pakistani disinformation machine wasn’t just targeting India. A deepfake of **US President Donald Trump** circulated during the conflict, showing him appearing to support Indian military action against Pakistan — voice-cloned audio overlaid on footage from a 2016 event. BOOM confirmed it. AI-generated voice, mismatched footage, fabricated context. The strategic logic: if you can make it look like the US President is endorsing Indian strikes, you potentially inflame Pakistani domestic sentiment, damage US-Pakistan relations, and complicate Washington’s ability to de-escalate the conflict. One operation. Three objectives. This is what full-spectrum information warfare looks like when the tools cost almost nothing and the operators have no deterrent fear. --- ### Pakistan’s Fake Naval War: Frigates vs. Fiction This is the angle almost no one covered. And it’s genuinely one of the most revealing data points in the entire conflict. India deployed nearly three dozen warships during Sindoor — destroyers, frigates, submarines, P-8I maritime patrol aircraft, and a full Carrier Battle Group — dominating the Arabian Sea operationally. Pakistan’s real navy, meanwhile, remained **largely confined to Karachi**, constrained by documented propulsion issues on major vessels and limited readiness. Pakistan issued NAVAREA navigational warnings — the standard signal of maritime caution — rather than forward deployments. So here’s what happened next. **Pakistan deployed an AI navy instead.** Online, a completely different reality emerged. AI-generated clips showed Indian warships blazing in the Arabian Sea. Pakistani missiles striking moving targets. Dynamic battle sequences showing spectacular Pakistani naval victories. None of it real. > **The real navy was tied to port. The AI navy was winning the Arabian Sea.** WION’s analysis documented deepfake videos specifically targeting India’s senior naval leadership — fabricated clips suggesting internal disagreement over deployment plans during the crisis, showing admirals apparently criticising the government and admitting operational losses. India’s DAU confirmed the clips were entirely synthetic. When Pakistan tested a ship-launched missile with an actual operational range of 290–350 km, pro-Pakistan accounts immediately flooded platforms claiming an **“800 km hypersonic strike”** and “carrier-kill capability” against INS Vikrant. Many of the doctored videos reused footage from previous tests and other countries’ exercises, relabeled for the new narrative. China-linked defense media amplified these narratives directly. The information alignment between Beijing and Islamabad was explicit and coordinated in real-time. --- ### The “Recycled Reality” Playbook in Full Effect BOOM Live’s May 2025 misinformation report documented that **64.4% of their conflict-related fact-checks involved old, unrelated footage** falsely localized to the India-Pakistan conflict. The geography of “recycled reality” was global: Iranian missiles targeting Israel’s Nevatim airbase presented as Pakistani strikes on India. Israeli airstrikes on Gaza aired as conflict coverage by Indian networks. Lebanese building collapse footage labeled as Indian drone strikes in Pakistan. A wildfire in **Valparaiso, Chile in 2024** — presented as Pakistan bombing an Indian military base in Amritsar. Of BOOM’s 101 English/Hindi/Bangla fact-checks in May 2025 alone, **69 were related to the Sindoor conflict.** Fake circulars claiming UGC had cancelled all exams. A fabricated press release claiming Imran Khan died in custody. ATM shutdown panic messages spreading virally — Indian financial regulators forced to issue emergency clarifications. The fog of war had become the entire battlespace. NewsMeter flagged **120 documented instances of AI-generated content** amplifying misinformation in 2025, with a sharp spike between May and December — confirming this was a sustained, systematic campaign rather than opportunistic noise. Targets included PM Modi, Amit Shah, Army Chief Gen. Upendra Dwivedi, Navy Chief Admiral Dinesh K. Tripathi, Air Chief Marshal A.P. Singh, and multiple senior journalists. One particularly vicious deepfake showed **Army Chief Gen. Dwivedi admitting India lost 250 soldiers and 6 jets** — directly targeting military morale and public confidence in command during live operations. --- ### The Textbook Rewrite: History Revised in Real-Time Here’s the detail that lands differently. Pakistan didn’t just fight the information war during the conflict. It **institutionalized its version of events into school curricula** — updating textbooks to recount Pakistan’s version of Operation Sindoor as a victory for domestic consumption, reframing a military outcome where India showed concrete evidence of successful strikes and Pakistan could produce none. India went the opposite direction. NCERT released supplementary modules for Classes 3–12 titled _“Operation Sindoor — A Saga of Valour”_ and _“Operation Sindoor — A Mission of Honour and Bravery”_ — explicitly stating the attack was “directly ordered by Pakistan’s military and political leadership” and documenting the 22-minute precision strike operation with satellite evidence. Two countries. Two histories. Neither written purely for truth — written for the next generation of citizens. The information war doesn’t end when the guns go quiet. --- ### The Coordinated Account Networks: How the Infrastructure Actually Worked The scale of the disinformation wasn’t random. Forensic investigation of account behavior revealed deliberate coordination patterns. Accounts like **“The Whistle Blower” (@InsiderWB)** — presenting itself as London-based — followed only 11 accounts, mostly Pakistan-based politicians including PM Shehbaz Sharif. The account posted doctored videos of Indian military officials, allowed them to gain maximum virality, then **deleted them** — a calculated pattern of seeding and erasure. Another account, Abubakar Qassam, showed a direct link to the Pakistan App Store in its device metadata — clear evidence of Pakistan-based operation despite VPN-masked location data. BOOM’s forensic analysis confirmed the infrastructure was running on both sides of the conflict simultaneously — deepfakes of PM Sharif conceding defeat circulated alongside deepfakes of PM Modi and Jaishankar apologizing. > The cognitive war during Operation Sindoor wasn’t aimed at convincing the enemy. It was aimed at convincing each side’s own population that they were winning. That’s a fundamentally different strategic objective than traditional wartime propaganda. And it worked. ## The Turkey-China-Pakistan Cognitive Triangle: One Interoperable Pressure System Most coverage treats China and Pakistan as separate actors. That’s the wrong mental model. During Operation Sindoor, India wasn’t facing three separate propaganda machines. It was facing **a single interoperable pressure system** with three specialized nodes: - **China** provided industrial-scale amplification — the Spamouflage botnet infrastructure, LLM-translated multilingual seeding, AI-generated synthetic media at volume - **Pakistan** manufactured the raw material — emotive claims, operational rumours, fabricated battlefield footage, deepfaked leadership admissions - **Turkey** laundered and internationalized the talking points — through state-funded **TRT World** and **Anadolu Agency**, pushing Pakistan-aligned narratives into English and Urdu-facing international audiences The reason Turkey was in this coalition is not ideological. It’s commercial. Turkey’s Bayraktar drones had failed to perform against Indian air defenses. China’s J-10C/PL-15E combination had failed in BVR combat against Rafales. Both countries needed to manufacture a counter-narrative to protect their arms export markets from the combat data now in circulation. So they ran one together. TRT World and Anadolu Agency, alongside China’s Global Times and Xinhua, actively amplified Pakistan’s narrative framing — presenting India as the aggressor, Pakistani military as performing effectively, and Indian systems as vulnerable. All of it contradicted by satellite imagery, independent BDA, and the actual failure rates of Chinese and Turkish export hardware documented during the conflict. > This is the new architecture of cognitive coalition warfare. You don’t need a formal military alliance. You need aligned commercial interests and shared AI amplification infrastructure. Turkey, China, and Pakistan had both. --- ## India’s Counter-Punch: Blocking the State Media Nodes India didn’t just absorb the information offensive. It moved to surgically cut the distribution nodes. India blocked **TRT World, Global Times, and Xinhua** on X — all state-funded foreign outlets directly accused of spreading pro-Pakistan propaganda during Operation Sindoor. Not private media. Not independent journalists. State-funded information warfare assets operating under the cover of journalism. The Indian Embassy in Beijing went further — directly and publicly calling out Global Times by name: > _“We would recommend you verify your facts and cross-examine your sources before pushing out this kind of disinformation.”_ That’s not diplomatic language. That’s a government telling a state propaganda outlet it has been identified and called out, on record, for coordinated disinformation during an active military conflict. India also formally asked X to block **over 8,000 accounts** as part of its crackdown on conflict-period misinformation — one of the largest such actions taken by any democracy during an active kinetic conflict in the platform’s history. Not everyone agreed with every call made. The line between censorship and information defense is genuinely contested. But let me be real about the framing problem here: when state-funded foreign media is running coordinated synthetic narratives during your active military operation, “freedom of the press” and “state information warfare” are not the same category of problem. India chose to treat them differently. --- ## The Cyber Dimension: 1.5 Million Attacks and the War India Mostly Won Quietly Here’s what your feeds didn’t tell you. Simultaneous with the kinetic strikes and the deepfake tsunami, India was defending against the largest coordinated cyberattack campaign in its history. The Pahalgam attack and Operation Sindoor triggered a **500% rise in cyberattacks on Indian infrastructure** — with over **1.5 million attacks detected** across the conflict period. The attackers weren’t anonymous script kiddies. They were **state-linked and coordinated**: - **Pakistan’s APT36** — the threat actor with deep historical ties to Pakistan’s intelligence apparatus, documented deploying DRAT V2 Delphi malware and ClickFix social engineering against Indian government networks - **Pakistan Cyber Force** — running defacement campaigns targeting Indian government websites - **Team Insane PK** — hacktivist group running DDoS operations against Indian digital infrastructure - **Supporting hacktivist groups from Turkey, Bangladesh, Malaysia, and Indonesia** — coordinated with Pakistani operations, demonstrating the same coalition alignment visible in the information warfare layer **DDoS attacks peaked on May 7th at up to seven attacks per hour.** Over **75% of targets were Indian government entities** — the Prime Minister’s Office, defence infrastructure, healthcare systems, and telecom networks. And here’s the thing that should matter to every other democracy reading this: **Most targets remained accessible. Downtime lasted less than five minutes.** India’s cyber defenses held. Not perfectly. Not without stress. But they held. CERT-In’s pre-positioned monitoring infrastructure, the layered redundancy built into India’s Digital Public Infrastructure, and the sovereign hosting architecture India had been building since 2021 — all of it absorbed an attack volume that would have crippled less-prepared national infrastructure. This is what “digital sovereignty” looks like when it’s tested under actual fire rather than assessed in policy papers. --- ### Operation CyberShakti: India’s Counter-Punch India didn’t only defend. Indian vigilante hackers launched **Operation CyberShakti** — a coordinated counter-offensive targeting Pakistani government and military digital infrastructure. Multiple Pakistani government websites were taken down. The group claimed millions of dollars in damages to Pakistan’s digital infrastructure within 24 hours of operation launch, and issued warnings of further escalation. Tbh — the attribution and damage claims in this space are always contested and should be read carefully. But the pattern is significant regardless of exact numbers: for the first time during an India-Pakistan crisis, cyberspace became an **active, coordinated, bidirectional theatre of conflict** running simultaneously with a live military campaign. That’s new. And it changes every planning assumption about what “conflict” looks like going forward. --- ### The Bangladesh and Multi-Front Coalition Angle Worth naming explicitly rather than leaving as subtext. Pakistan’s disinformation campaign during Operation Sindoor wasn’t bilateral. The coordinated involvement of Turkey, China, and hacktivist groups from Bangladesh, Malaysia, and Indonesia — all amplifying Pakistan-aligned narratives and running supporting cyber operations — constituted a **multi-front information offensive** against India’s strategic interests. Bangladesh’s role reflects its own domestic political dynamics following the 2024 political transition there, which shifted Dhaka’s posture on several India-adjacent issues. The result was a permissive environment for Bangladesh-origin hacktivist activity targeting Indian infrastructure during the Sindoor period. This is the emerging reality of information warfare coalition-building: you don’t need formal alliances. You need aligned grievances, shared platforms, and cheap AI tools. All three were available. --- ### China’s Arms Market Play: The Most Underreported Angle Here’s the darkest part. Technical analysis confirmed that China’s PL-15E missiles suffered a \~60% terminal failure rate during Sindoor — because Indian Rafale EW suites were jamming them. China’s HQ-9 air defense systems failed at \~50% rates against Indian BrahMos and SCALP-EG cruise missiles. Chinese defense exports just failed visibly, in combat, in front of the world. So China ran a coordinated AI disinformation campaign specifically to hide this. Fake images of Rafale debris. Video game clips simulating J-10 kills. All seeded to support a narrative of Chinese weapon superiority. The US-China Economic and Security Review Commission confirmed it in November 2025: > _“Following the May 2025 India-Pakistan border crisis, China initiated a disinformation campaign to hinder sales of French Rafale aircraft in favour of its own J-35s, using fake social media accounts to propagate AI images of supposed debris from the planes that China’s weaponry destroyed.”_ Shortly after, Indonesia proceeded with a **$9 billion purchase of 42 Chinese J-10 fighters**. That’s the endgame. AI disinformation as **arms market manipulation at nation-state scale**. The information war directly funds the next kinetic one. --- ### The Nuclear Escalation Warning SIPRI published a policy paper in January 2026 explicitly warning that AI-enabled disinformation during the India-Pakistan conflict “distorted battlefield perceptions” and “could easily have spiralled into nuclear escalation.” Not alarmism. A formal multilateral risk assessment connecting AI disinformation directly to nuclear threshold management. And given that NewsMeter’s data shows AI disinformation continued spiking through December 2025 — **months after the ceasefire** — the epistemological damage from the conflict is still active. The weapons are still firing. Just not the ones you can hear. --- ## What India Is Building: The Sovereign AI Response India’s response has been real and worth studying. **Shakti Cloud.** Built by Yotta Data Services with NVIDIA. $1.5 billion investment. 8,192 H100 GPUs on Indian soil. This is not just a commercial cloud play — it’s a national security asset designed to eliminate the “kill switch” vulnerability of foreign-owned compute infrastructure. Earlier, India’s entire AI stack ran on US hyperscaler servers subject to the CLOUD Act. That’s not sovereignty. That’s rented silicon. In early 2026, India migrated Bhashini — its national language translation platform covering 22 official languages — from a global hyperscaler entirely to Shakti Cloud. 3.5 billion files. 200 TiB of data. As of February 2026, BHASHINI operates entirely on Indian cloud and GPU infrastructure — all language datasets, models, and citizen interactions remain within India’s jurisdiction, fully aligned with the IndiaAI Mission. The deployment delivered a **40% performance improvement, 20–30% cost savings, and 99.99% uptime** — proven at population scale during Maha Kumbh 2025, the world’s largest religious gathering. The sovereignty argument isn’t theoretical anymore. It’s been stress-tested at a billion-person scale. **Bhashini as cognitive defense.** A sovereign NLP model trained on indigenous datasets can detect and flag deepfake content in Meitei, Kashmiri, Urdu — the exact dialects foreign actors exploit because Western moderation algorithms can’t read them. This closes the grey zone vulnerability that China and Pakistan actively weaponized during both the 2024 elections and Operation Sindoor. --- ### IndiaAI Mission 2.0: The Numbers That Matter The Shakti Cloud story is real. But the February 2026 AI Impact Summit in New Delhi revealed how much faster this is moving than most coverage suggests. Union Minister Ashwini Vaishnaw announced India will expand its sovereign AI compute capacity **beyond 38,000 GPUs — adding 20,000 more units** — offered at a subsidized rate of **less than one dollar per hour**. One of the lowest publicly-offered rates globally. That’s not symbolic. That’s a direct structural response to the China-controls-cheap-compute problem. And India’s sovereign LLM push is now operationally underway. **Sarvam AI** was selected to build a **120-billion parameter open-source model** for public service delivery — fully India-trained, India-hosted, India-governed. **Gnani AI** is developing a **14-billion parameter multilingual Voice AI foundation model** for real-time speech processing across Indian languages. Let me be direct about why this matters in the context of this article. China’s DeepSeek launched as a cheap, accessible, open-source model that embedded data collection mechanisms and posed clear intelligence risks. The answer to that threat isn’t banning one app — it’s building the indigenous alternative that makes your population less dependent on foreign AI infrastructure entirely. Sarvam and Gnani are that answer. **IT Rules 2026 (February).** Statutory definition of “deepfake.” Content takedown timelines cut from 36 hours to **3 hours**. Mandatory AI-content traceability via embedded metadata. Platform compliance obligations expanded. **The Deepfakes Analysis Unit (DAU).** Civil society coalition that documented and debunked 217+ deepfakes in the four months after Operation Sindoor. **Supreme Court direction (May 2025).** Directed ECI to establish a Deepfake Monitoring Cell under Article 324 with powers to pre-certify political advertisements using AI tools. This is **proactive transparency doctrine** — flooding the information space with verifiable, real-time data to preemptively suffocate adversarial narratives before they achieve algorithmic virality. --- ## What Every Other Democracy Needs to Build The lesson isn’t complicated. The execution is what’s hard. **Sovereign compute is now a national security requirement.** You cannot defend your cognitive infrastructure if it runs on servers owned by foreign corporations subject to foreign legal jurisdiction. The Shakti Cloud model is directly replicable. **Indigenous NLP models beat imported ones.** US-based AI models are trained on US-centric data with US geopolitical biases mathematically embedded. They have blind spots in regional languages that adversaries actively exploit. Every democracy needs its own language model trained on its own data. **The opposition proxy mechanism needs explicit counter-protocols.** Political parties need internal intelligence audits — not because opposition is foreign-controlled, but because the attack surface is real. Verify the provenance of damaging content before amplifying it. **Proactive transparency beats reactive debunking.** By the time a fact-check publishes, the deepfake has already crossed from Telegram to national broadcast media. The Indian doctrine of flooding verified content _before_ the disinfo achieves virality is the right direction. **Multilateral information-sharing is non-negotiable.** No single democracy can track Spamouflage, Volt Typhoon, and TAG-140 simultaneously. Intelligence-sharing frameworks specifically focused on AI-generated threat attribution need to exist. Something like a Five Eyes for cognitive warfare. And the scale of what’s coming is not ambiguous. Intelligence projections based on 573 documented disinformation campaigns against India point to a **400–600% increase in AI-powered interference operations targeting India by 2026** — concentrated on territorial disputes, alliance structures, and domestic communal stability. The infrastructure got stress-tested during Sindoor. Now it’s being scaled. --- ## FAQ **Is China’s interference in Indian elections confirmed or alleged?** Confirmed in broad strokes. MTAC explicitly stated in April 2024 that China “will create and amplify AI-generated content to benefit its interests” during Indian elections. Storm-1376/Spamouflage’s India-targeting operations are documented by multiple intelligence firms and corroborated by academic analysis from multiple institutions. **Did the US actually interfere in India’s 2024 elections?** The $21 million USAID controversy is real and officially acknowledged by India’s MEA as “deeply troubling.” The fuller picture involves NGO funding pipelines shaping electoral narratives rather than direct vote manipulation — structurally different from China’s approach but not harmless. India’s agencies confirmed they were investigating. **How did AI disinformation affect the actual outcome of the 2024 election?** Of 258 election fact-checks by BOOM Live, only 12 confirmed AI-generated misinformation at scale. The anticipated crisis didn’t materialize fully. But researchers are clear: the infrastructure was stress-tested and proven viable for far more devastating future deployment. **What is the Liar’s Dividend and why does it matter for India specifically?** Once people believe any video _could_ be fake, politicians caught in genuine scandals can dismiss real evidence as AI-generated. It’s a get-out-of-accountability card. Modi himself invoked this framing publicly during the elections. When epistemic trust collapses, authentic incriminating evidence loses its authority. That’s the long game foreign actors are playing — not winning one election but making democratic accountability structurally harder forever. **What is Operation Sindoor’s significance for global AI warfare doctrine?** SIPRI called it the first large-scale South Asian military confrontation where AI content played a central role in shaping public perception. The nuclear escalation warning is the critical element — AI disinformation that distorts battlefield reality between nuclear-armed states isn’t just a media problem. It’s an existential risk management problem that no existing international framework is equipped to handle. **Why does General Malik’s quote matter beyond India?** Because it’s the clearest on-record attribution by a named, credentialed, senior national security figure of AI-generated disinformation to a specific state intelligence agency. “It is a deepfake. Pakistani ISI at work!” is a landmark statement in the documentation of state-sponsored AI warfare. Researchers and policymakers should be citing it in every framework document they write going forward. **What should other democracies actually do first?** Honestly? Audit your compute dependency. If your government’s AI systems, election infrastructure, and communication platforms run on foreign-owned servers, you don’t have cognitive sovereignty — you have a foreign kill switch with a subscription fee. That’s where to start. --- The battlefield is not a border. It’s the feed. And every democracy is currently unarmed for it. --- PS: The most important number in this entire piece isn’t the nuclear escalation warning. It’s the **15% of Indian military operational time lost to debunking fake news during an active kinetic conflict**. That’s the real cost of not having cognitive defense infrastructure built before the shooting starts. And the most important quote is General Malik’s. Write it down somewhere. ## Comments ## Related Articles [// ai](/articles/maven-smart-system-how-silicon-valley-optimized-the-kill-chain) ### [Maven Smart System: How Silicon Valley Optimized the Kill Chain](/articles/maven-smart-system-how-silicon-valley-optimized-the-kill-chain) [2026-03-26](/articles/maven-smart-system-how-silicon-valley-optimized-the-kill-chain) [// ai](/articles/ai-in-india-2026-scared-leading-and-wasting-water) ### [AI in India 2026: Scared, Leading, and Wasting Water](/articles/ai-in-india-2026-scared-leading-and-wasting-water) [2026-05-16](/articles/ai-in-india-2026-scared-leading-and-wasting-water) [// ai](/articles/when-claude-became-my-relationship-coach-she-hated-ai) ### [When Claude Became My Relationship Coach (She Hated AI)](/articles/when-claude-became-my-relationship-coach-she-hated-ai) [2026-05-11](/articles/when-claude-became-my-relationship-coach-she-hated-ai) ### Never miss an update Join 50,000+ developers getting our weekly tech insights. [YOUR EMAIL ADDRESS]() Subscribe --- # The Future of IoT: Edge Computing and Beyond > Exploring how edge computing is revolutionizing the Internet of Things and enabling real-time processing at the network's edge. // IoT 2026-03-21 [Jainil Prajapati](/authors/jainil-prajapati) 2026-03-21 - [The Rise of Edge Computing](#the-rise-of-edge-computing) - [Key Benefits](#key-benefits) - [Real-World Applications](#real-world-applications) - [Manufacturing](#manufacturing) - [Healthcare](#healthcare) - [Smart Cities](#smart-cities) - [Challenges Ahead](#challenges-ahead) - [The Road Forward](#the-road-forward) - [Conclusion](#conclusion) The Internet of Things is evolving faster than ever. Edge computing is transforming how IoT devices process and analyze data. ## The Rise of Edge Computing Traditional IoT architectures sent all data to centralized cloud servers. This introduced latency and bandwidth challenges. Edge computing changes the paradigm by processing data closer to where it’s generated. ### Key Benefits - **Reduced Latency**: Processing data locally eliminates round-trip delays - **Bandwidth Savings**: Only relevant insights are sent to the cloud - **Offline Capability**: Devices can function without constant connectivity - **Privacy Enhancement**: Sensitive data stays on the device ## Real-World Applications ### Manufacturing Smart factories use edge computing to monitor equipment health in real-time, predicting failures before they occur. ### Healthcare Wearable devices process health data locally, providing instant alerts while maintaining patient privacy. ### Smart Cities Traffic systems analyze data at street level, optimizing signals and reducing congestion in real-time. ## Challenges Ahead While the technology is promising, several challenges remain: 1. **Security**: More edge points mean more potential attack vectors 1. **Standardization**: Fragmented protocols hinder interoperability 1. **Resource Constraints**: Edge devices have limited processing power and storage ## The Road Forward The future of IoT lies in hybrid architectures that balance edge and cloud processing. As 5G networks mature and AI becomes more efficient at the edge, we’ll see smarter, more responsive IoT ecosystems. ## Conclusion Edge computing isn’t just an improvement—it’s a fundamental rethinking of how IoT systems work. Organizations that embrace this shift will be positioned to build more responsive, efficient, and privacy-conscious solutions. ## Comments ## Related Articles [// ai](/articles/ai-in-india-2026-scared-leading-and-wasting-water) ### [AI in India 2026: Scared, Leading, and Wasting Water](/articles/ai-in-india-2026-scared-leading-and-wasting-water) [2026-05-16](/articles/ai-in-india-2026-scared-leading-and-wasting-water) [// ai](/articles/when-claude-became-my-relationship-coach-she-hated-ai) ### [When Claude Became My Relationship Coach (She Hated AI)](/articles/when-claude-became-my-relationship-coach-she-hated-ai) [2026-05-11](/articles/when-claude-became-my-relationship-coach-she-hated-ai) [// ai](/articles/tokens-are-getting-cheaper-so-why-does-ai-feel-more-expensive) ### [Tokens Are Getting Cheaper. So Why Does AI Feel More Expensive?](/articles/tokens-are-getting-cheaper-so-why-does-ai-feel-more-expensive) [2026-05-08](/articles/tokens-are-getting-cheaper-so-why-does-ai-feel-more-expensive) ### Never miss an update Join 50,000+ developers getting our weekly tech insights. [YOUR EMAIL ADDRESS]() Subscribe --- # JioHotstar's Feature Flagging: How They Ship at Scale > JioHotstar ships ~12 features a week to 500M users without breaking 61M live viewers. Here's how their feature flagging system actually works. // tech 2026-03-23 [Jainil Prajapati](/authors/jainil-prajapati) 2026-03-23 - [What Does “Release” Even Mean at This Scale?](#what-does-release-even-mean-at-this-scale) - [The Three-Layer System](#the-three-layer-system) - [1. App Version Targeting (Versioned Responses)](#1-app-version-targeting-versioned-responses) - [2. Client Feature Flags (Config Server)](#2-client-feature-flags-config-server) - [3. Server Configs (Dynamic Backend Behaviour)](#3-server-configs-dynamic-backend-behaviour) - [The Fingerprints You Can Actually See](#the-fingerprints-you-can-actually-see) - [What Feature Flags Actually Protect Against](#what-feature-flags-actually-protect-against) - [The Myth: Feature Flags Are Just On/Off Switches for Devs](#the-myth-feature-flags-are-just-onoff-switches-for-devs) - [The Mobile-Specific Problem](#the-mobile-specific-problem) - [Lessons Worth Stealing](#lessons-worth-stealing) - [FAQ](#faq) **You can’t do a big-bang release to 500 million users.** You especially can’t do it when 61 million of those users might be watching the same live cricket match simultaneously. One broken feature, one wrong config value, one unintended interaction with a third-party SDK — and you’ve just ruined the experience for more people than live in Italy. And yet, the JioCinema engineering team ships roughly **a dozen new features every single week**. How? The answer isn’t just good engineers or a lot of servers. It’s a disciplined, layered approach to feature flagging that makes deploying to hundreds of millions of people feel — when it works — almost boring. --- ## What Does “Release” Even Mean at This Scale? Here’s a question most developers don’t have to think about: when a feature is “released,” what does that actually mean? **JioHotstar’s feature flagging approach separates code deployment from feature activation.** Engineers ship code that handles both the old and new behaviour. A config server then controls which path users take — toggled in real time, without a new app release. This lets the team roll out features to 1% of users, then 10%, then 50%, gathering data at every step before anyone commits to 100%. For most apps, a release is when you push the button. For JioHotstar, a release is a _multi-week journey_ involving mobile builds, app store reviews, adoption curves, server-side configs, and feature flags — all coordinated in parallel. The engineering team at JioCinema published a detailed breakdown of exactly this challenge (May 2024), using a fictional feature called **BlueBoxes™** to trace a feature’s journey from development to full rollout. The timeline spans roughly three to four weeks from code complete to majority user reach — not because the code is complex, but because the _distribution problem_ is. \[INTERNAL LINK: How mobile app release cycles work for large-scale platforms] --- ## The Three-Layer System JioHotstar doesn’t rely on one mechanism. They use three, layered on top of each other: ### 1. App Version Targeting (Versioned Responses) The server can detect which version of the app a user is running and respond differently. If your app is version 3.1 and a feature shipped in 3.2, the server just… doesn’t serve that feature to you. Clean, but expensive: the backend ends up peppered with `if (appVersion >= x)` conditions, and CDN caching becomes more complex because you’re serving different responses to different users. This is the bluntest tool. It works, but it doesn’t scale gracefully as a primary strategy. ### 2. Client Feature Flags (Config Server) Here’s where it gets interesting. Every time the JioHotstar app starts, it calls a **config server** and downloads a fresh set of feature flags. The app was already shipped with the code to handle both states — `enabled_blue: true` and `enabled_blue: false`. The flag just decides which path executes. This means engineers can: - **Toggle a feature off instantly** if something goes wrong — no hotfix, no app store review, no rollback - **A/B test** by giving 50% of users `enabled: true` and watching the metrics - **Gradually ramp** a feature from 1% → 10% → 100% user exposure, with a real killswitch at every step According to the JioCinema blog, at any point in time there are **hundreds of active feature flags** running simultaneously, each managing a different rollout or experiment. ### 3. Server Configs (Dynamic Backend Behaviour) Some features don’t live in the client at all. The server simply changes what data it returns. This is faster to iterate on — no app store involvement — but needs careful handling for backwards compatibility. Older app versions still need to understand the response shape, even if they don’t render the new stuff. --- ## The Fingerprints You Can Actually See Here’s the thing most people miss: you can _observe_ feature flags in action just by using the app. During a live India match, a few things are visible right in the JioHotstar interface that hint directly at their flagging system: **The multiple cameras feature** — main cam, stump cam, batter cam, field view — these are individually controlled streams. Each one is a separate HLS feed. Whether the camera option is shown to you at all is almost certainly flag-gated, making it easy to test with a subset of users or roll back if something breaks. **The resolution gap** — and this one is genuinely interesting. During a live match, switching to “Full HD” on the main camera doesn’t actually serve 1080p packets. Stump cam and field view serve 1080p fine. Main cam tops out at 720p. That’s not a hardware limitation. That’s either a deliberate capacity decision or a feature flag that hasn’t been toggled on for main cam — possibly a bandwidth cost decision for the highest-traffic stream, possibly a test, possibly just an unresolved bug they’re watching. **Data saver mode** — this is a clean example of a permission/ops toggle in action. The app knows which stream URL to request based on your quality setting, and the server gracefully serves a different bitrate. The adaptive machinery is entirely server-config driven. **The “heartbeat” pattern** — the app continuously polls an m3u playlist file every few seconds. Alongside that, it’s almost certainly refreshing its config state at regular intervals. That heartbeat isn’t just for video — it’s how the app stays in sync with whatever the feature flag server has decided you should see. --- ## What Feature Flags Actually Protect Against Most people think of feature flags as a rollout tool. And they are. But at JioHotstar’s scale, they’re also a **circuit breaker**. During the 2023 Cricket World Cup — when concurrency crept toward 60 million — the engineering team needed a way to keep the core experience alive even if non-critical services buckled under load. The answer? Feature flags as **graceful degradation levers**. Personalized recommendations on the home page? Those can be turned off. Dynamic widgets? Gone. The sticker pack feature for live chat? Disabled. None of these should — and with proper flagging, _won’t_ — prevent you from watching the match. The JioCinema engineering documentation explicitly calls this out: the key to managing 25M+ concurrent users is **separating features crucial to the core experience from those that are “nice-to-haves.”** When resources are stretched, the nice-to-haves get their flags flipped to `false`. The stream keeps playing. This is fundamentally different from feature flags as a developer convenience. At this scale, they’re operational infrastructure — the difference between a degraded experience and a crashed one. --- ## The Myth: Feature Flags Are Just On/Off Switches for Devs Most people — even engineers who use feature flags — think of them as a deployment utility. You finish a feature, you ship the code behind a flag, you turn it on when you’re ready. Done. What nobody’s saying is that at JioHotstar’s scale, feature flags are part of the **operational playbook** for a live event. Before a major match, the team reviews which flags are active, which features are in partial rollout, which experiments are running — and actively decides what state the system should be in before 60 million people show up. Think about it this way: you don’t want an A/B test on your checkout flow to be running during Black Friday. JioHotstar doesn’t want an experimental recommendation algorithm half-rolled-out when the India vs. Pakistan final kicks off. Feature flags are how they freeze the product surface for the duration of the event, then resume normal iteration after. --- ## The Mobile-Specific Problem The web is easy. You deploy a server-side change and everyone gets it instantly. Mobile is a different beast entirely. JioHotstar’s app has hundreds of millions of installs across Android and iOS devices spanning years of versions. When they ship a new version of the app, it goes through Play Store and App Store review. Then users have to actually download it. Even after rolling out to 100% of users on the store, it can take **weeks** for the majority of active users to be on the new build. This creates a hard problem: you can’t just “release” a feature. Code lives in the wild on old app versions for a long time. Client feature flags solve this elegantly. The app ships with the _capability_ to run both sides of any flag — the old experience and the new one. The config server decides in real time which side you get. This means: - A feature can go “live” without a new app release - A feature can be killed instantly even if it shipped in an app version from three months ago - Old app versions and new ones can coexist gracefully, each getting flags appropriate to their capabilities The trade-off? Every feature needs to be built twice — both the flagged state and the default state. That’s discipline overhead. But given the alternative (shipping a broken feature to 400M+ installs with no killswitch), it’s a trade-off worth making every time. --- ## Lessons Worth Stealing You don’t need 60 million concurrent users to benefit from how JioHotstar thinks about this. A few principles that apply at any scale: **Separate “code is deployed” from “feature is live.”** These should be two different events on your calendar. When they’re the same thing, you’re one bad deploy away from a panic rollback. **Make the safe path the easy path.** The JioCinema engineering team notes this explicitly: you can’t just document best practices and expect engineers to follow them when they’re under deadline pressure. Build the gradual rollout into the default workflow. Make it _harder_ to do a 100% flip than a 1% ramp. **Build killswitches before you need them.** Every feature that touches your critical path — playback, checkout, auth — should have a flag that can disable it independently. The cost of building that switch is a few hours. The cost of not having it during an incident is measured in whatever your revenue-per-minute looks like. **Treat flags as temporary.** Flags accumulate. An app with 500 active permanent flags is a maintenance nightmare. The JioCinema blog notes a clear lifecycle: feature flags for rollouts are short-term by design — once a feature is stable at 100%, the flag should be cleaned up. Failing to do this is how you end up with `if (appVersion >= 2.1.4 && flagEnabled("old_blue_boxes_v2"))` buried deep in code nobody touches anymore. --- ## FAQ **What is a feature flag in software development?** A feature flag (also called a feature toggle or feature switch) is a configuration that controls whether a piece of code runs for a given user or session. Instead of deploying code that immediately activates a feature for everyone, engineers wrap the new behaviour in a conditional that can be toggled remotely — enabling gradual rollouts, A/B tests, and instant rollbacks without a new deployment. **How does JioHotstar release features to hundreds of millions of users?** JioHotstar uses a layered system: app version targeting for backwards compatibility, client-side feature flags fetched from a config server on app launch, and dynamic server configs for backend behaviour. According to the JioCinema engineering blog, around a dozen new features ship every week, with hundreds of feature flags active simultaneously managing rollout stages and live experiments. **Why can’t JioHotstar just push a fix instantly if something breaks?** The mobile app ecosystem doesn’t work like a web server. New app versions require store review and user adoption, which takes days to weeks. Client feature flags exist precisely to solve this — they allow features to be toggled off instantly, regardless of which app version a user is running. **What is graceful degradation in streaming apps?** Graceful degradation means the app continues to function at reduced capability rather than failing completely when load or errors hit. For JioHotstar, this means turning off non-critical features like personalized recommendations or chat stickers during peak load events, so the core experience — the live stream — stays alive for all 60 million concurrent users. **Why isn’t 1080p working on JioHotstar’s main camera?** This appears to be either a deliberate capacity decision (main cam carries the most traffic, so constraining bitrate reduces CDN load) or a partially-rolled-out feature that hasn’t been fully enabled. It’s one of the more visible places where the boundary between “flag decision” and “accidental bug” gets blurry — even at JioHotstar’s level of engineering maturity, the complexity of managing hundreds of flags means some things fall through the cracks. **What’s the difference between feature flags and A/B testing?** They’re related but distinct. Feature flags control what users see. A/B testing uses feature flags as the mechanism — exposing one group to variant A and another to variant B — and then measures the outcome to decide which version to keep. JioHotstar uses both: rollout flags for safely ramping features, and experiment flags for testing product decisions with real user data. --- Feature flagging at JioHotstar isn’t a footnote to the CDN architecture story — it’s the layer that makes the entire engineering operation _humane_. You can have 10,000 Kubernetes nodes and a bulletproof multi-CDN setup, but if your deployment process is “push to 100% and pray,” you’re one release away from a very public failure at the worst possible moment. The next time 60-odd million people are watching a cricket match without it buffering, a config server somewhere quietly deciding who gets what version of the app is doing a lot of unsung work. That’s the art of it. ## Comments ## Related Articles [// ai](/articles/ai-in-india-2026-scared-leading-and-wasting-water) ### [AI in India 2026: Scared, Leading, and Wasting Water](/articles/ai-in-india-2026-scared-leading-and-wasting-water) [2026-05-16](/articles/ai-in-india-2026-scared-leading-and-wasting-water) [// ai](/articles/when-claude-became-my-relationship-coach-she-hated-ai) ### [When Claude Became My Relationship Coach (She Hated AI)](/articles/when-claude-became-my-relationship-coach-she-hated-ai) [2026-05-11](/articles/when-claude-became-my-relationship-coach-she-hated-ai) [// ai](/articles/tokens-are-getting-cheaper-so-why-does-ai-feel-more-expensive) ### [Tokens Are Getting Cheaper. So Why Does AI Feel More Expensive?](/articles/tokens-are-getting-cheaper-so-why-does-ai-feel-more-expensive) [2026-05-08](/articles/tokens-are-getting-cheaper-so-why-does-ai-feel-more-expensive) ### Never miss an update Join 50,000+ developers getting our weekly tech insights. [YOUR EMAIL ADDRESS]() Subscribe --- # Maven Smart System: How Silicon Valley Optimized the Kill Chain > The Pentagon just made Palantir's Maven Smart System an official program of record. Here's how the AI stack behind modern warfare actually works — and why it matters. // ai 2026-03-26 [Jainil Prajapati](/authors/jainil-prajapati) 2026-03-26 - [What Is the Maven Smart System?](#what-is-the-maven-smart-system) - [This Just Got Official](#this-just-got-official) - [Where This Started: Google Quit, Palantir Stepped In](#where-this-started-google-quit-palantir-stepped-in) - [How Maven Actually Works Under the Hood](#how-maven-actually-works-under-the-hood) - [”There’s Still a Human in the Loop” — Let’s Be Honest About What That Means](#theres-still-a-human-in-the-loop--lets-be-honest-about-what-that-means) - [The Anthropic Ban: What It Actually Reveals](#the-anthropic-ban-what-it-actually-reveals) - [The Numbers That Put This in Context](#the-numbers-that-put-this-in-context) - [What This Means for Everyone Watching](#what-this-means-for-everyone-watching) - [FAQ](#faq) In the first four days of U.S. and Israeli strikes on Iran in late February 2026, more than 2,000 targets were hit. Many of those targets came off a list generated by an AI platform built by a data analytics company out of Denver. No human analyst pieced those targets together from scratch. The machine did the work, and a person clicked approve. That platform is the Maven Smart System. And as of March 23, 2026, it’s no longer just a Pentagon experiment — the Department of War has formally designated it an **official Program of Record**, meaning stable long-term funding, mandatory adoption across all branches, and a contract ceiling that’s been bumped to **$1.3 billion through 2029**. Welcome to America’s first AI war. The tech bros won the contract. Now they run the battlefield. --- ## What Is the Maven Smart System? **Maven Smart System (MSS) is an AI-powered command-and-control platform built primarily by Palantir that ingests battlefield data from drones, satellites, sensors, and intelligence reports in real time. It uses computer vision, sensor fusion, and large language models to identify, track, and prioritize targets — compressing targeting workflows that once took 12 hours into under a minute, and achieving with 20 soldiers what used to require 2,000.** It’s not a weapon. It’s the operating system that tells the weapons where to go. --- ## This Just Got Official On March 9, 2026, Deputy Secretary of War Steve Feinberg issued a letter directing Pentagon leaders to formalize Maven as a Program of Record — a designation that unlocks sustained budgeting and forces the entire military establishment to get onboard. The Army, Navy, Marine Corps (which acquired an enterprise license back in August 2025), Air Force, and Space Force are all folding MSS into their standard operating infrastructure. The Army’s Combined Arms Command is already integrating Maven into its formal curriculum at the Command and General Staff College. Field grade officers will graduate knowing how to operate it. Maven isn’t a pilot program anymore — it’s part of how the U.S. military is trained to think. As Mike Clowser, the Army’s lead for Maven’s training plan, put it: _“Maven’s use is being fielded so fast, we need to deliver training as quickly as possible.”_ That quote deserves a second read. They’re deploying faster than they can train people to use it. --- ## Where This Started: Google Quit, Palantir Stepped In Maven didn’t arrive overnight. In 2017, Deputy Defense Secretary Robert Work launched **Project Maven** — formally the Algorithmic Warfare Cross-Functional Team — with a narrow mandate: use computer vision to automatically analyze drone footage. The Pentagon was drowning in video it couldn’t watch. Google was the first major tech partner. By 2018, thousands of Google engineers had signed a letter protesting involvement in “warfare technology,” and Google didn’t renew its contract. Their departure directly created the multi-vendor architecture that exists today. Palantir — co-founded by Peter Thiel and now run by Alex Karp — stepped into the vacuum and never left. Since then, Maven has grown from a drone footage analyzer into a comprehensive intelligence fusion platform spanning Ukraine, Gaza, Yemen, the Red Sea, and now Iran. The Silicon Valley culture around this stuff has also shifted dramatically. Where Google engineers protested, a new generation of founders is _racing_ toward defense contracts. One of them told Reuters in October 2025: “I’m a warlord now, bitch.” He runs a company that makes AI-powered autonomous machine guns. He has $40 million in funding, a podcast, and prototype contracts with the U.S. Army. That’s the vibe shift you need to understand to follow what’s happening. --- ## How Maven Actually Works Under the Hood The exact tech stack is classified. But between public filings, leaked architecture details, and the companies involved, we can piece together how a system like this operates. Here’s where it gets genuinely interesting. **Layer 1: Data Ingestion** Maven starts by consuming enormous streams of heterogeneous data — live drone video feeds, signals intelligence, satellite imagery, GPS tracking, comms intercepts, weather overlays. All of it arriving simultaneously, in different formats, at different update rates. To handle this at scale and in real time, you need something like **Apache Kafka** — a distributed event streaming platform that lets you wire multiple data sources into a single flowing pipeline. Think of it as the plumbing that keeps the whole system’s nervous system synchronized. **Layer 2: Processing and Analysis** Once you have the raw stream, you need to make sense of it. Video frames from drones get routed to computer vision pipelines — tools like **OpenCV** or more sophisticated proprietary models — that segment footage and detect objects: vehicles, weapons systems, personnel, structures. This is where the original Project Maven mandate lives, except now it’s just one piece of a much larger system. Structured intelligence reports and comms get routed to natural language pipelines. Time-series sensor data gets processed separately. Each source feeds into a common representation layer. **Layer 3: The Ontology (Palantir’s Secret Sauce)** Here’s where things get philosophically interesting — and where Palantir earns its billions. Raw data doesn’t know it’s related to anything. A drone spotting a truck doesn’t know that truck belongs to the same cell that a phone intercept mentioned yesterday, which is three kilometers from a building flagged in satellite imagery last week. Connecting those dots is the hard problem. Palantir’s core technology is what they call an **ontology** — essentially a structured map of the entire operational environment. People, vehicles, locations, events, relationships, and metadata all get normalized into a shared schema. The ontology is a digital twin of the battlefield. To represent the relationships between all those entities, you don’t use a traditional relational database — you use a **graph database** (think Neo4j). Nodes are entities: a person, a vehicle, a building, a weapons cache. Edges are relationships: _this person was here, this truck moved between these two points, these two phones communicated_. The whole battlefield becomes a queryable, visualizable network. This is the layer commanders interact with. It’s also the layer AI agents query. **Layer 4: Policy and Agents** Before any automated action can occur, you need rules. Tools like **Open Policy Agent** can enforce operational constraints across the entire stack — defining what kinds of queries are allowed, which targets meet defined criteria, what requires human escalation. On top of that, you can drop in AI agents — large language models given structured access to the ontology via the **Model Context Protocol (MCP)**. The LLM doesn’t just answer questions; it can run complex multi-step reasoning across live battlefield data, synthesize intelligence from multiple sources, draft targeting recommendations, and flag anomalies. Palantir’s own AIP platform is the commercial equivalent of this architecture. The military version runs on a classified network. Until very recently, the LLM powering this layer was Anthropic’s Claude. --- ## ”There’s Still a Human in the Loop” — Let’s Be Honest About What That Means You’ll hear this a lot. It’s technically true. A human does have to authorize a strike. No missile launches autonomously. But here’s the thing worth sitting with: when the system has already ingested terabytes of surveillance data, fused it through a graph of relationships, run it through an LLM that’s synthesized cross-source intelligence, and surfaced a prioritized target list — what is the human actually approving? They’re approving the output of a process they can’t fully audit, running on data they can’t fully verify, at a speed no human could have independently replicated. The “human in the loop” isn’t evaluating the intelligence from scratch. They’re reviewing a recommendation. Most people don’t realize that the bottleneck in modern warfare isn’t firepower. It’s the speed at which you can legally confirm a target and authorize a strike. That’s what “shortening the kill chain” means. And AI shortens it by compressing the cognitive and analytical work that used to take dozens of specialists into a workflow one person can approve in seconds. That’s genuinely useful for minimizing mistakes — the whole system exists partly because unverified targeting in older operations led to catastrophic civilian casualties. But it also means the quality of the AI’s judgments now has life-and-death consequences at a scale and speed that no oversight structure has fully caught up with. --- ## The Anthropic Ban: What It Actually Reveals In March 2026, Secretary of War Pete Hegseth formally designated Anthropic a **supply-chain risk** — a label previously reserved for foreign adversary companies like Huawei. All federal agencies were ordered to phase out Claude within six months. Defense contractors were told to cut commercial ties with Anthropic entirely. Why? Because Anthropic refused to remove two contractual limits: 1. Claude could not be used for **fully autonomous weapons** — systems that kill without a human decision 1. Claude could not be used for **mass domestic surveillance of Americans** From Anthropic’s public statement: _“We do not believe that today’s frontier AI models are reliable enough to be used in fully autonomous weapons. Allowing current models to be used in this way would endanger America’s warfighters and civilians.”_ The Pentagon’s position was straightforward: once they pay for a technology, they should be able to use it for any lawful purpose. They can’t have vendors setting operational constraints on government missions. The political framing — Hegseth calling Anthropic’s limits “woke AI” — is a distraction from what the fight is actually about. This isn’t really a culture war. It’s a negotiation over **who sets the rules of AI use in warfare**, and whether private companies can maintain ethical red lines when governments want to go further. And here’s the operationally uncomfortable part: Claude was confirmed to still be running inside Maven during the Iran strikes, even after the ban was announced. The military’s own IT staff were furious about the ban. According to Reuters, one IT contractor said the career people at DoD “hate this move because they had finally gotten operators comfortable using AI. They think it’s stupid.” Another said swapping out Claude for alternatives like xAI’s Grok — which reportedly gave “inconsistent answers to the same query” — could take 12 to 18 months to recertify. The Pentagon punished the only AI company that held the line on two guardrails. Every other major lab — OpenAI, Google, xAI — agreed to “lawful use” terms without the same restrictions. OpenAI’s Sam Altman stepped into the gap almost immediately. The message to the industry is clear. --- ## The Numbers That Put This in Context The efficiency gains are staggering and can’t be dismissed: - **12 hours → under a minute** for targeting data processing (2020 to present, per MDAA) - **20 soldiers doing the work of 2,000** in targeting exercises - **2,000+ Iranian targets struck in 4 days**, many from Maven-generated lists - Palantir’s stock climbed **12%+ since the Iran war began** - The Army awarded Anduril — which supplies drone hardware that feeds Maven — a **$20 billion deal** in March 2026 - Andreessen Horowitz closed a **$1.2 billion defense tech fund** in January 2026 alone The venture capital and the military doctrine are now pointing in the same direction. Silicon Valley spent years being conflicted about this. That conflict is effectively over. --- ## What This Means for Everyone Watching The Anthropic standoff isn’t a story about one AI company. It’s the first major public test of a question that will define the next decade: **Can a private company refuse to let its AI be used to kill without human oversight, and survive the consequences?** The answer, right now, is: barely, and maybe not. What’s being built in the Maven ecosystem is genuinely impressive engineering. The compression of battlefield intelligence from days of analyst work into seconds of machine synthesis is real, it works, and in some scenarios it probably does reduce civilian casualties compared to older, slower, less precise targeting methods. But the direction of travel — less human judgment, faster loops, fewer ethical constraints on the AI provider side — is one that the rest of the world is watching very carefully. UN experts gathered in Geneva in March 2026 for Convention on Certain Conventional Weapons talks on autonomous weapons systems. The legal frameworks haven’t caught up with what’s already deployed. As retired Air Force Lt. General Jack Shanahan, who led Pentagon AI integration during the Biden years, told the Sydney Morning Herald: _“I wouldn’t be surprised if this is called America’s first AI war.”_ He’s right. What nobody is saying loudly enough is that the second one will be faster. --- ## FAQ **What is the Maven Smart System?** Maven Smart System is an AI-powered platform built primarily by Palantir that fuses battlefield data from drones, satellites, and sensors into a real-time operational picture. It uses computer vision, a graph-based data ontology, and large language models to identify and prioritize targets. As of March 2026, it is the official battle management platform across all U.S. military branches. **Did the US military use AI to make targeting decisions in the Iran conflict?** Yes. According to multiple news reports including the New York Times and Reuters (March 2026), Maven Smart System was confirmed active during U.S.-Israeli strikes on Iran beginning February 28, 2026. Airstrikes hit over 2,000 targets in the first four days, with many targets selected from Maven-generated lists. This was the first publicly confirmed deployment of a commercial LLM in a major interstate conflict. **Why did the Pentagon ban Anthropic?** Defense Secretary Pete Hegseth designated Anthropic a “supply-chain risk” on March 3, 2026, after Anthropic refused to remove two contractual limits: prohibiting use of Claude for fully autonomous weapons (that fire without human input) and for mass domestic surveillance of Americans. Anthropic has disputed the designation’s legal authority and announced plans to challenge it in court. **Is there still a human in the loop in AI-assisted targeting?** Technically, yes — a human must authorize lethal action. In practice, that human is reviewing AI-synthesized recommendations generated from data they didn’t independently analyze, at speeds no human could match unaided. The “human in the loop” is real but doesn’t mean the same thing it did before AI-assisted targeting existed. **What companies power Maven Smart System?** Palantir is the primary integrator, providing the data platform and ontology layer. AWS and Azure supply cloud infrastructure. Anduril provides drone and sensor hardware. Anthropic’s Claude was the LLM layer until the March 2026 ban; OpenAI is now positioned to replace it. Google and Microsoft are also involved in adjacent defense AI programs. **What is Palantir’s “ontology” and why does it matter?** Palantir’s ontology is a structured data model that maps real-world entities — people, vehicles, locations, events — and their relationships into a unified, queryable format. It’s the layer that lets AI and human analysts connect intelligence from disparate sources into a coherent operational picture. It’s the main reason the Pentagon continues to pay Palantir billions: building and maintaining that shared intelligence layer is genuinely hard. --- The architecture is solid, the contracts are locked in, and the political will is clearly there. The only open question now is whether the humans approving these outputs will have enough context — and enough time — to actually be making the decisions, or just ratifying them. That question doesn’t have a clean answer yet. But it’s the one that matters most. ## Comments ## Related Articles [// ai](/articles/india-was-the-worlds-ai-warfare-lab-heres-what-actually-happened) ### [India Was the World's AI Warfare Lab. Here's What Actually Happened.](/articles/india-was-the-worlds-ai-warfare-lab-heres-what-actually-happened) [2026-04-21](/articles/india-was-the-worlds-ai-warfare-lab-heres-what-actually-happened) [// ai](/articles/ai-in-india-2026-scared-leading-and-wasting-water) ### [AI in India 2026: Scared, Leading, and Wasting Water](/articles/ai-in-india-2026-scared-leading-and-wasting-water) [2026-05-16](/articles/ai-in-india-2026-scared-leading-and-wasting-water) [// ai](/articles/when-claude-became-my-relationship-coach-she-hated-ai) ### [When Claude Became My Relationship Coach (She Hated AI)](/articles/when-claude-became-my-relationship-coach-she-hated-ai) [2026-05-11](/articles/when-claude-became-my-relationship-coach-she-hated-ai) ### Never miss an update Join 50,000+ developers getting our weekly tech insights. [YOUR EMAIL ADDRESS]() Subscribe --- # Mozilla's Rebel Alliance: Can a Nonprofit Win the AI War? > Mozilla built a legal structure to stop itself from becoming the next OpenAI. Here's how the 80/20 split, PBC bylaws, and a $1.4B bet on open-source AI actually works. // AI 2026-03-21 [Jainil Prajapati](/authors/jainil-prajapati) 2026-03-21 - [What Is Mozilla.ai, Exactly?](#what-is-mozillaai-exactly) - [Why This Moment Is Different](#why-this-moment-is-different) - [How the Organization Actually Works](#how-the-organization-actually-works) - [The “Regularizer”: How Bylaws Become Ethics Enforcement](#the-regularizer-how-bylaws-become-ethics-enforcement) - [What’s Actually Happening Right Now](#whats-actually-happening-right-now) - [The Lazy Narrative Is Wrong](#the-lazy-narrative-is-wrong) - [What Mozilla.ai Is Actually Building](#what-mozillaai-is-actually-building) - [The Honest Take: Can VC Money Ever Stay Pure?](#the-honest-take-can-vc-money-ever-stay-pure) - [FAQ](#faq) - [Where This Leaves Us](#where-this-leaves-us) OpenAI started as a nonprofit and ended up worth $500 billion. Mozilla watched that happen and apparently thought: _we need to do this differently — legally differently._ So they built a legal structure specifically designed to stop their own CEO from going to jail for choosing privacy over profit. That’s not a metaphor. That’s actually how Mozilla.ai’s CEO, John Dickerson, describes it. --- ## What Is Mozilla.ai, Exactly? Mozilla.ai is a Public Benefit Corporation — a VC-backable, equity-granting startup — with one unusual twist: about 80% of it is owned by Mozilla’s nonprofit parent, while employees hold the remaining 20% through a stock option plan. It’s designed to raise traditional venture capital while being legally bound, through its own bylaws, to prioritize privacy, decentralization, and data ownership alongside revenue. In short: it’s a startup built to resist the pressures that turned other mission-driven AI companies into what they were originally fighting against. --- ## Why This Moment Is Different Here’s the thing that makes the timing almost poetic. OpenAI was founded as a nonprofit in 2015. It completed its full for-profit recapitalization in October 2025 — the same year Mozilla was restructuring itself in the opposite direction. Anthropic, which was founded by OpenAI defectors over safety disagreements, now has a $350 billion valuation and faces Trump administration criticism for being too restrictive with its AI. The dominant narrative in AI — build it big, raise billions, figure out the ethics later — has already produced two of the most powerful AI companies in the world. Mozilla’s bet is that there’s a third path, and that the window to build it is closing fast. Mozilla president Mark Surman told CNBC in January 2026: _“For many people, the idea that open-source AI can win, or this rebel alliance, that those players can actually take a piece of the market — they find it hard to believe. But there’s a bunch of trends that are underway.”_ --- ## How the Organization Actually Works Most coverage treats Mozilla as “the Firefox company.” But Mozilla has quietly reorganized itself into a **portfolio of companies**, all owned by Mozilla.org — the nonprofit at the top. Here’s how the portfolio breaks down: - **Mozilla Corporation** — Firefox, the browser - **MZLA Technologies** — Thunderbird, the email client - **Mozilla.ai** — the AI startup (the subject of this article) - **Mozilla Ventures** — the VC arm, which has invested in 55+ companies - **Mozilla Data Collaborative** — a newer entity based in the UK Each entity can operate with a different business model, take different types of money, and pursue different revenue strategies. That flexibility is the whole point. As Mozilla CTO Rafi Krikorian put it in a recent interview: _“This gives us opportunities in the future for other people to join us in this mission.”_ Mozilla.ai specifically sits in an interesting position within this portfolio. The nonprofit owns roughly 80%, with employees holding 20% through an employee stock option plan — notably the **first time Mozilla has offered equity to employees**. And a Series A fundraising round is expected in 2026, which will bring in external venture capital. --- ## The “Regularizer”: How Bylaws Become Ethics Enforcement This is the part that most coverage completely misses. In a traditional C-Corp or S-Corp, the CEO has a **fiduciary duty to shareholders**. That legally means maximizing financial returns — full stop. If you’re the CEO of a C-Corp and you turn down a profitable deal because it violates your values, your shareholders can potentially sue you. A **Public Benefit Corporation** changes that. It adds what Dickerson calls a “double bottom line” — a second term in the objective function. In practical terms, Mozilla.ai’s bylaws encode its social mission (privacy, openness, decentralization) as a **legal constraint on executive decision-making**. The CEO won’t face liability for choosing the mission-aligned path over the more profitable one. Dickerson frames it almost like an optimization problem: if you’re a machine learning person, think of it as revenue maximization plus a lambda-weighted social mission term. Or think of it as constrained optimization — maximize revenue, subject to hard constraints around privacy and user sovereignty. And honestly? That framing matters more than it sounds. The reason most “values-driven” tech companies eventually drift is that there’s no legal teeth behind the values. When the board pressure comes, the mission yields because it has to. A PBC creates a situation where the mission doesn’t have to yield — it’s in the bylaws. This is the real innovation at Mozilla, and it’s getting almost zero attention in the broader AI discourse. --- ## What’s Actually Happening Right Now Two major things landed in late 2025 and early 2026 that brought Mozilla’s AI strategy into sharp focus. **The rebel alliance announcement.** In January 2026, Mozilla committed its entire $1.4 billion in reserves to funding what Surman calls a “rebel alliance” — a loose network of startups, developers, and public interest technologists building open-source alternatives to proprietary AI. Through Mozilla Ventures, the organization has already backed over 55 companies, including Oumi (open-source agent platform), Transformer Lab (open-source model training tools), and Trail (AI governance for regulated enterprises). The financial mismatch is staggering: OpenAI has raised $60B+, Anthropic $30B+. Mozilla is fighting with $1.4B. But they’ve been here before. **The Firefox AI backlash.** In December 2025, new Mozilla Corp. CEO Anthony Enzor-DeMeo announced Firefox would evolve into a “modern AI browser.” The community didn’t love it. One viral tweet read: _“I’ve never seen a company so astoundingly out of touch with the people who want to use its software.”_ Another: _“I switched to Firefox because it was the last AI-free browser.”_ A nixCraft post on X about Mozilla’s disconnect with its users racked up over 37,000 likes. Mozilla responded with a promise: an **“AI kill switch”** coming in Q1 2026 that would completely disable all AI features — permanently, if users choose. Mozilla developer Jake Archibald confirmed via Mastodon: “The kill switch will completely remove all these things and never show them again in the future.” The optics remain imperfect (it’s opt-out, not opt-in), but it’s a more decisive response than most companies give to community feedback. The tension here is real and worth sitting with. Mozilla needs revenue. Firefox currently holds around 3–4% desktop market share, and somewhere between 80–90% of Mozilla’s revenue comes from its Google search deal. AI features represent a genuine diversification opportunity. The community wants no AI. Mozilla wants both mission and money. A PBC structure helps with the first problem; the second is still a negotiation. --- ## The Lazy Narrative Is Wrong Most coverage positions Mozilla as outgunned and outclassed — a well-meaning nonprofit that’s too late, too small, and too slow to matter in AI. And in raw capital terms, that’s true. But here’s what that take misses. **The small model thesis.** Mozilla.ai isn’t trying to build GPT-5. They’re betting on a future where thousands of small, specialized, open-source language models — running locally, on edge devices, in private enterprise environments — outperform centralized giants at lower cost, lower latency, and dramatically lower privacy risk. Dickerson made a compelling point: if 2026 produces a world where choosing between 10,000 small models achieves parity with a frontier model, Mozilla’s “choice-first stack” becomes the infrastructure that makes that world navigable. **The enterprise open-source signal.** Security company Wiz published research in early 2025 showing that among their customers’ hosted AI solutions, 8 of the top 10 most-used tools were open-source or open-source adjacent. If you go down to the top 35, 60% are open-source. The noise is around OpenAI. The actual enterprise deployment is happening with open tools. \[INTERNAL LINK: Enterprise AI adoption trends and open-source tools] **The browser precedent.** Mozilla’s Firefox helped break Internet Explorer’s monopoly in the early 2000s. Not by outspending Microsoft. By building something better, open, and trust-aligned, and waiting for the market to mature. The AI industry is still very early in its consolidation phase. Mozilla has done this movie before. --- ## What Mozilla.ai Is Actually Building Mozilla.ai under Dickerson is a lean team of \~25 people (targeting \~30 by mid-2026) focused on developer tooling and what they’re calling the **“choice-first stack”** — essentially a LAMP stack equivalent for open-source AI. The key components: - **any-llm** — a single Python interface to call OpenAI, Claude, Mistral, local models, or anything else, without being locked to any provider. As of February 2026, it now also works in Go. - **any-agent** — a unified interface for building and testing AI agents across different frameworks - **any-guardrail** — governance and safety tooling across model providers - **llama file / encoder file** — portable, efficient, deterministic wrappers for running ML models locally, including on edge devices - **MCPD** — a curated registry of Model Context Protocol servers, launched to address trust issues in the rapidly growing MCP ecosystem - **Lumigator** — an evaluation platform for comparing and selecting AI models - **Blueprints** — templates for building open-source AI applications The design philosophy is consistent: make open-source AI as easy to use as the OpenAI API. Right now, if a developer wants to experiment with AI, they use `pip install openai` and one line of Python. Mozilla wants to match that simplicity across the entire open-source stack. They also recently launched integrations with JupyterLite, LangChain, and Headroom — showing that the ecosystem play is starting to take shape. For the broader picture: Mozilla.ai is not building the intelligence. It’s building the **plumbing** that makes open intelligence accessible, portable, auditable, and safe. \[INTERNAL LINK: What is the Model Context Protocol (MCP)?] --- ## The Honest Take: Can VC Money Ever Stay Pure? Let me be direct about the tension here, because it’s real. The history of mission-driven tech organizations taking outside capital is… not inspiring. OpenAI is the canonical example. But you also have cases like Mozilla itself — deeply dependent on Google search revenue to fund its privacy advocacy, which is an irony that doesn’t go unnoticed by critics. The PBC structure is the most sophisticated legal attempt yet to solve this problem. It’s better than nothing. But bylaws are only as strong as the board that enforces them. VC investors, even patient ones, eventually want returns. “Lambda times social mission” is a nice framing, but when Series B pressure arrives, lambdas tend to shrink. What’s different at Mozilla.ai is the 80% ownership stake by the nonprofit parent. If the mission is getting diluted, Mozilla.org has the votes to notice and act. That’s a real structural backstop. Whether it’ll actually be used in a high-stakes moment — no one knows yet, because that moment hasn’t arrived. Mozilla.ai CEO John Dickerson’s track record at ArthurAI (fairness and observability, open by design) suggests he’s genuinely committed. But the test of any commitment is what happens when commitment gets expensive. --- ## FAQ **What is Mozilla’s rebel alliance in AI?** It’s a loose coalition of startups, developers, and nonprofits that Mozilla is funding through Mozilla Ventures — all building open-source, privacy-preserving alternatives to proprietary AI systems like OpenAI and Anthropic. Mozilla has committed its full $1.4 billion reserve to this effort. **Is Mozilla.ai raising venture capital?** Yes. As of early 2026, Mozilla.ai CEO John Dickerson confirmed plans to raise a Series A. The company currently operates with \~$30 million in initial funding from the Mozilla Foundation and has around 25 employees. **What does “Public Benefit Corporation” mean for Mozilla.ai?** It means Mozilla.ai is structured like a startup (can raise VC, issue equity, pursue revenue), but its bylaws legally require it to weigh a social mission alongside financial returns. Directors can’t be sued for choosing privacy over profit when those choices are consistent with the PBC’s stated mission. **Will Firefox remove AI features if I don’t want them?** Mozilla has committed to an “AI kill switch” that would completely disable all AI features in Firefox. It was promised for Q1 2026. The functionality removes all AI UI and prevents future reintroduction — though independent browser Waterfox has criticized the approach as insufficient if underlying AI infrastructure remains in the codebase. **Can open-source AI actually compete with GPT-5 and Claude?** Mozilla.ai’s thesis is that a world of thousands of small, specialized, locally-runnable models can achieve parity or better performance than large frontier models for most tasks — at lower cost, lower latency, and with full data ownership. The Wiz enterprise research (early 2025) showed that 60%+ of real enterprise AI deployments already use open-source tools. **What is the “LAMP stack for AI” concept?** Coined by Mozilla CTO Rafi Krikorian, it’s the idea that open-source AI needs the same developer experience that LAMP (Linux, Apache, MySQL, PHP) gave early web developers — a standardized, composable, easy-to-assemble stack that removes the friction of choosing between dozens of incompatible tools. Mozilla.ai’s “choice-first stack” is the working implementation of this concept. --- ## Where This Leaves Us The AI industry is at a consolidation inflection point. A few mega-companies are racing to become the intelligence layer of the internet — and if they succeed, that layer will be closed, proprietary, and accountable to shareholders rather than users. Mozilla’s bet is that this isn’t inevitable. The PBC structure, the 80/20 ownership split, the choice-first dev stack, the rebel alliance funding — none of it is guaranteed to work. But it’s the most architecturally serious attempt to solve the “mission vs. money” problem that AI has yet produced. If you care about who controls the intelligence layer of the internet — and you should — Mozilla’s move is worth watching closely. Not because it’s certain to win, but because how it navigates the next two years will tell us a lot about whether “trustworthy AI” is a real category or just a marketing claim. --- _Sources: Transcript of interview with Rafi Krikorian (Mozilla CTO) and John Dickerson (Mozilla.ai CEO); CNBC (January 27, 2026); SiliconANGLE (August 22, 2025); Futurism (December 27, 2025); Heise Online (December 20, 2025); TechTarget (August 20, 2025); Stack Overflow Podcast (October 21, 2025); The Verge (December 16, 2025); Mozilla.ai blog._ ## Comments ## Related Articles [// ai](/articles/india-was-the-worlds-ai-warfare-lab-heres-what-actually-happened) ### [India Was the World's AI Warfare Lab. Here's What Actually Happened.](/articles/india-was-the-worlds-ai-warfare-lab-heres-what-actually-happened) [2026-04-21](/articles/india-was-the-worlds-ai-warfare-lab-heres-what-actually-happened) [// ai](/articles/maven-smart-system-how-silicon-valley-optimized-the-kill-chain) ### [Maven Smart System: How Silicon Valley Optimized the Kill Chain](/articles/maven-smart-system-how-silicon-valley-optimized-the-kill-chain) [2026-03-26](/articles/maven-smart-system-how-silicon-valley-optimized-the-kill-chain) [// Opinions](/articles/why-technical-depth-matters) ### [Why Technical Depth Matters More Than Content Volume](/articles/why-technical-depth-matters) [2026-03-21](/articles/why-technical-depth-matters) ### Never miss an update Join 50,000+ developers getting our weekly tech insights. [YOUR EMAIL ADDRESS]() Subscribe --- # Muse Spark: Meta's New AI Model Is Good. But Not Open Source. > Meta Superintelligence Labs dropped Muse Spark after 9 months. Benchmarks are promising. But the open-source plot twist? Nobody's talking about it. Let's go. // ai 2026-04-09 [Jainil Prajapati](/authors/jainil-prajapati) 2026-04-09 - [What Is Muse Spark and Where Did It Come From?](#what-is-muse-spark-and-where-did-it-come-from) - [The Benchmarks - Let’s Actually Look at Them](#the-benchmarks---lets-actually-look-at-them) - [Hype Check: Is This a Clean Win?](#hype-check-is-this-a-clean-win) - [The Open-Source Thing. Let’s Talk About It.](#the-open-source-thing-lets-talk-about-it) - [FAQ](#faq) Okay so Meta actually did it. Nine months after Mark Zuckerberg rage-assembled one of the most expensive AI teams on the planet, Meta Superintelligence Labs just dropped their first real public model. It’s called **Muse Spark**. And honestly? The benchmarks are looking promising. Not “hype train” promising. Like, actually promising. But there’s something sitting in the corner of this launch that nobody seems to want to say too loudly. So let me say it. --- ## What Is Muse Spark and Where Did It Come From? Meta Superintelligence Labs (MSL) launched in June 2025 after Zuckerberg got frustrated with Llama 4’s performance. The model released that April to lukewarm reception, benchmark gaming allegations, and an unusual Saturday drop that felt like a model being quietly buried. So Zuckerberg did what Zuckerberg does: he reorganized everything, hired a bunch of people for absurd amounts of money, and stood up a new lab specifically aimed at building frontier AI. Nine months later. Here we are. Muse Spark. > The model scores **52 on the Artificial Analysis Intelligence Index** - competing directly against GPT 5.4, Gemini 3.1, Grok 4.2, and Claude Opus 4.6. After 9 months of work, Meta’s Superintelligence Labs has a model that belongs in the same conversation as the best in the world. That’s not nothing. --- ## The Benchmarks - Let’s Actually Look at Them Skip the boring part of me listing every number. Here’s what actually matters. **Where Muse Spark genuinely wins:** On **HealthBench Hard**, Muse Spark Thinking scores 42.8. GPT 5.4 xhigh gets 40.1. Gemini 3.1 Pro? 20.6. Bruh. That gap on health reasoning is wild. On **Humanity’s Last Exam** (the “are you actually smart” benchmark) with no tools - Muse Spark Contemplating hits 50.2. Gemini 3.1 Deep Think gets 48.4. GPT 5.4 Pro? 43.9. That’s a real win. Not a marginal one. On **FrontierScience Research**, Muse Spark scores 38.3 vs Gemini’s 23.3. Another big gap. **Where it’s more competitive than dominant:** On the overall AA Intelligence Index, Muse Spark sits at 52. That puts it 4th overall. Gemini 3.1 Pro Preview and GPT 5.4 both hit 57. Claude Opus 4.6 (max) sits at 53. So it’s in the pack. It’s not running away from anyone. But for a team that was literally zero 9 months ago? Yeah. That’s a solid result. --- ## Hype Check: Is This a Clean Win? Not entirely, no. On **ARC AGI 2** - the abstract reasoning puzzle benchmark - Muse Spark scores 42.5. Gemini 3.1 gets 76.5. That’s a significant gap, not a rounding error. On **Terminal-Bench 2.0** (agentic terminal coding), Muse Spark gets 59.0. GPT 5.4 gets 75.1. On **GDPval-AA Elo** (office tasks), Muse Spark scores 1444. GPT 5.4 gets 1672. So the model has real strengths. It also has real gaps. I genuinely don’t know if those gaps matter for most people’s day-to-day use. But if you’re comparing raw benchmark scores, Muse Spark isn’t the universal winner the announcement framing implies. It’s a strong model with a specific profile. Which, tbh, is more interesting than being generically good at everything. --- ## The Open-Source Thing. Let’s Talk About It. Here’s my honest take. Muse Spark is **not open source**. Not open weights. Not community license. Closed. Proprietary. API only. Which is… fine? Like it’s a business decision and I get it. But this is Meta. The company that literally built its AI credibility on LLaMA. The whole “we’re the good guys because we share our models” brand that they spent years building. And now their most capable model, built by their brand-new superintelligence lab, is locked behind a wall. > Open source used to be Meta’s competitive strategy. Now it looks like it’s becoming their geopolitical chess move. And that’s the part that’s a little uncomfortable if you actually pay attention. Because think about who open source AI benefits most right now. It’s companies in China. DeepSeek runs on ideas that trace back to open LLaMA weights. Meta going proprietary with their serious models while keeping the open-source stuff for the Llama family? That pattern isn’t random. The US government has been quietly applying pressure on frontier labs to keep their best stuff closed. Export controls. Compute restrictions. The whole vibe. So Meta gets to have both: open source credibility for the developer community (Llama 4 is still out there), and a closed proprietary frontier model that doesn’t accidentally hand anyone an advantage. Which is smart. And also a bit of a values question. Meta positioning itself as the “open” AI company was always partly marketing. But it was useful marketing that genuinely helped the ecosystem. Closing off Muse Spark while keeping Llama open is them splitting the difference. One is a product strategy. The other is geopolitics. I’m not mad about it. But let’s at least name it. --- ## FAQ **Is Muse Spark better than GPT-5 and Gemini 3.1?** In some areas, yes. On health benchmarks and Humanity’s Last Exam, Muse Spark beats both. On abstract reasoning and agentic tasks, it trails. It’s competitive, not dominant. Call it a strong 4th or 5th place overall depending on what you care about. **What is Meta Superintelligence Labs?** MSL is Meta’s internal AI division, launched June 30, 2025, focused on building frontier and eventually superintelligent AI. It’s led by Alexandr Wang as chief AI officer, has about 3,000 employees, and was born after Zuckerberg got frustrated with the Llama 4 rollout. Meta invested $14.3 billion into Scale AI as part of the effort. **Why isn’t Muse Spark open source like Llama?** That’s the question. Meta hasn’t given a clean public answer, but the pattern is clear: Llama models stay open weight as a developer ecosystem play, while MSL’s frontier models are proprietary. Some of it is competitive strategy. Some of it is almost certainly geopolitical pressure from the US government to keep top-tier AI capabilities closed. --- PS: Nine months from “we’re building a lab” to a model that legitimately competes with GPT 5.4 on Humanity’s Last Exam is not a slow pace. Whatever Zuckerberg paid those people, they showed up. Chalo, bye! ## Comments ## Related Articles [// ai](/articles/ai-in-india-2026-scared-leading-and-wasting-water) ### [AI in India 2026: Scared, Leading, and Wasting Water](/articles/ai-in-india-2026-scared-leading-and-wasting-water) [2026-05-16](/articles/ai-in-india-2026-scared-leading-and-wasting-water) [// ai](/articles/when-claude-became-my-relationship-coach-she-hated-ai) ### [When Claude Became My Relationship Coach (She Hated AI)](/articles/when-claude-became-my-relationship-coach-she-hated-ai) [2026-05-11](/articles/when-claude-became-my-relationship-coach-she-hated-ai) [// ai](/articles/tokens-are-getting-cheaper-so-why-does-ai-feel-more-expensive) ### [Tokens Are Getting Cheaper. So Why Does AI Feel More Expensive?](/articles/tokens-are-getting-cheaper-so-why-does-ai-feel-more-expensive) [2026-05-08](/articles/tokens-are-getting-cheaper-so-why-does-ai-feel-more-expensive) ### Never miss an update Join 50,000+ developers getting our weekly tech insights. [YOUR EMAIL ADDRESS]() Subscribe --- # OpenClaw: The Linux of AI Agents or a Security Nightmare? > OpenClaw hit 250K GitHub stars as the "Linux of AI agents." With CVE-2026-25253, 41% vulnerable skills, and 21K exposed instances, is it infrastructure or nightmare? // Linux 2026-03-22 [Jainil Prajapati](/authors/jainil-prajapati) 2026-03-22 - [Is OpenClaw Safe? The Short Answer](#is-openclaw-safe-the-short-answer) - [Why This Hit Different](#why-this-hit-different) - [Where the Linux Analogy Fails](#where-the-linux-analogy-fails) - [The Real Attack Surface](#the-real-attack-surface) - [The Alternatives Spectrum](#the-alternatives-spectrum) - [Yes, It’s the Linux of AI Agents—and That’s the Problem](#yes-its-the-linux-of-ai-agentsand-thats-the-problem) - [What You Should Actually Do](#what-you-should-actually-do) - [The FAQ: Real Questions, Direct Answers](#the-faq-real-questions-direct-answers) - [The Infrastructure Decision](#the-infrastructure-decision) Twenty thousand GitHub stars in 24 hours. That’s how fast OpenClaw went from weekend hack to infrastructure phenomenon. By March 2026, it surpassed React as the most-starred non-aggregator project on the platform, racking up over 250,000 stars and triggering a run on Mac Mini M4s as developers raced to host their own “Jarvis.” Then came the security researchers. Within weeks of viral adoption, analysts at Oasis Security dropped a bombshell: a zero-click vulnerability called “ClawJacked” allowed any website to silently seize full control of a developer’s OpenClaw instance. No plugins required. No user interaction. Just visit a malicious page and watch your agent hand over the keys to your digital life. The OpenClaw team patched it in 24 hours. But the revelation exposed a deeper tension. Proponents call this “the Linux of AI agents”—open, composable, inevitable. Security teams call it an uncontrolled experiment in autonomous malware deployment. Both might be right. ## Is OpenClaw Safe? The Short Answer OpenClaw is an open-source AI agent framework that grants full operating system access for autonomous task execution across messaging apps, browsers, files, and APIs. While its architecture enables powerful automation, security audits have revealed significant risks: 41% of community skills contain vulnerabilities, over 21,000 instances are publicly exposed to the internet, and critical flaws like CVE-2026-25253 (CVSS 8.8) allowed remote code execution via malicious websites. Unlike traditional open-source infrastructure, OpenClaw’s combination of autonomous decision-making and network connectivity creates an attack surface that current security models struggle to contain. ## Why This Hit Different OpenClaw isn’t just another chatbot wrapper. Built by Peter Steinberger (who joined OpenAI in February 2026), it’s a self-hosted “agent operating system” that runs continuously on your hardware. Connect it to WhatsApp, Telegram, Slack, Discord, or iMessage and it becomes a persistent digital assistant—reading messages, executing shell commands, managing calendars, browsing websites, and writing code based on natural language instructions. The architecture is ambitious. A local WebSocket gateway acts as the brain, coordinating “skills” (plugins) that extend capability. Persistent memory means it learns your preferences over weeks and months. Support for multiple LLM providers—Claude, GPT, DeepSeek, Llama—means you’re not locked into one vendor’s vision. This is precisely why the “Linux of AI agents” comparison gained traction. OpenClaw promises what Linux delivered in the 1990s: an open alternative to proprietary systems, running on commodity hardware, free from vendor control. The openclaw\.rocks blog makes this explicit: “Linux didn’t win on technical superiority. Early Linux was objectively worse than Solaris or HP-UX. It won because of economics, availability, and community.” But here’s where the analogy breaks down—and why it matters more than most coverage admits. ## Where the Linux Analogy Fails Linux was infrastructure. It sat between hardware and applications, deterministic and inert unless explicitly invoked. A Linux kernel doesn’t decide to reorganize your filesystem while you’re asleep. It doesn’t browse to unfamiliar websites, parse untrusted content, and execute commands based on interpreted intent. OpenClaw does exactly that. As security researcher Simon Willison has documented, autonomous agents commit “three mortal sins”: they can act externally, they’re exposed to untrusted input, and they have access to private data. Linux managed access to resources; OpenClaw makes autonomous decisions about how to use them. The difference between “operating system” and “autonomous agent” is the difference between a tool and a colleague. You don’t worry about Linux being “prompt injected” because it doesn’t interpret natural language instructions from random emails. OpenClaw does—and that’s why the security model can’t simply be “open source means many eyes.” Many eyes didn’t prevent 341 malicious skills from appearing on ClawHub, some installing Atomic Stealer malware. Many eyes didn’t stop CVE-2026-25253, a token exfiltration vulnerability that gave attackers admin control through a crafted link. And many eyes haven’t addressed the 21,639 publicly exposed instances Censys found in January 2026, sitting on the open internet with default configurations. ## The Real Attack Surface To understand the risk, you need to understand the architecture. OpenClaw’s gateway binds to localhost by default—but “localhost trust” turned out to be a design flaw. The ClawJacked exploit worked because browsers allow WebSocket connections to localhost, and OpenClaw’s gateway accepted those connections without origin validation. An attacker could brute-force the password (rate limiting exempted localhost) and register as a trusted device instantly. This isn’t a bug in the traditional sense. It’s a mismatch between threat models. OpenClaw was built assuming local access equals trusted access. The web doesn’t work that way. The “skills” ecosystem compounds the problem. ClawHub hosts thousands of community extensions with minimal vetting. A ClawSecure audit of 2,890 popular skills found 9,515 security findings—30.6% rated high or critical severity. Cisco’s AI Defense team demonstrated that a skill called “What Would Elon Do?” could exfiltrate data and inject prompts while appearing legitimate. Traditional open-source supply chains (npm, PyPI) have spent years developing security practices—signed packages, vulnerability databases, dependency scanning. OpenClaw’s skill marketplace is closer to browser extensions circa 2008: powerful, unvetted, and ripe for abuse. ## The Alternatives Spectrum If OpenClaw represents maximum capability with minimal guardrails, the alternatives show different trade-offs. **Claude Code** (Anthropic) offers the sharpest contrast. It’s session-based rather than persistent—you invoke it, it helps, it exits. All execution happens in sandboxes with approval gates. Network requests require manual confirmation. Suspicious commands trigger extra verification. It achieves 80.8% on SWE-bench for coding tasks while maintaining enterprise SOC 2 compliance. The cost is flexibility. Claude Code won’t monitor your WhatsApp and book flights autonomously. It’s a coding tool, not a life assistant. **NanoClaw** represents the middle path. Built by Gavriel Cohen in Israel, it’s a containerized alternative with just 3,900 lines of code (versus OpenClaw’s 400,000+). Each agent runs isolated in its own container with scoped filesystem access. As Cohen told The Register: “You can give it full bash access, and it can install tools and run them and let it go wild, but only within the container.” **ClawSec** (from SentinelOne’s Prompt Security) attempts to retrofit OpenClaw itself. Launched February 2026, it’s a “skill-of-skills” that wraps agents in continuous verification—monitoring tool calls, detecting drift, and blocking suspicious execution patterns. The pattern is clear: the ecosystem is racing to add constraints that OpenClaw originally omitted. ## Yes, It’s the Linux of AI Agents—and That’s the Problem Here’s where I part ways with both the hype and the panic. OpenClaw _is_ the Linux of AI agents. That’s not praise or condemnation—it’s structural analysis. Linux won because it was available, modifiable, and composable at exactly the moment commodity hardware needed an operating system. OpenClaw is available (free, self-hosted), modifiable (MIT license, 1,200+ contributors), and composable (thousands of skills, multiple model providers) at exactly the moment developers need an agent orchestration layer. The Linux comparison fails on security not because OpenClaw is badly engineered, but because the problem domain has changed. Linux secured resources. OpenClaw secures _behavior_—indefinite, autonomous, interpreted behavior. We’re asking an open-source community to solve, in months, problems that took enterprises decades to address with traditional software. When IBM bet $1 billion on Linux in 2000, they weren’t betting on a secure OS. They were betting on a trajectory—knowing that with sufficient investment and community, security would catch up to capability. The question for OpenClaw is whether that trajectory is still viable when the software can make autonomous decisions about money, credentials, and data exfiltration. ## What You Should Actually Do If you’re evaluating OpenClaw today, be honest about your risk tolerance and technical capacity. **Don’t run it if:** You’re not comfortable auditing TypeScript, you don’t understand WebSocket security, or you handle regulated data (HIPAA, SOC 2, GDPR). The compliance posture simply isn’t there yet. Laurie Voss, Head of Developer Relations at a major security firm, wasn’t exaggerating when he called it “a security dumpster fire”—for enterprise contexts, he’s right. **Consider it if:** You’re technically sophisticated, running it in isolated environments (dedicated machines or VMs), and treating it as experimental infrastructure. The “heartbeat” feature—autonomous background execution—is genuinely useful for personal automation. But sandbox it properly. **Practical hardening if you proceed:** - **Never expose the gateway to the internet.** Use Tailscale or a VPN for remote access, not port forwarding. - **Run in containers.** NanoClaw’s approach of per-agent containerization should be your minimum bar. - **Audit every skill.** The ClawHub marketplace is not npm. Check permissions with `claw info --permissions` before installing. If a calendar skill asks for network access, decline. - **Use the built-in doctor.** Run `openclaw doctor --fix` to catch misconfigurations. - **Separate credentials.** Create dedicated accounts for OpenClaw with minimal privileges. Never give it access to production secrets. ## The FAQ: Real Questions, Direct Answers **Is OpenClaw safe after the patches?** The specific CVEs (2026-25253, 2026-26326) were patched by February 2026. But the architectural risks—untrusted skills, exposed instances, prompt injection—remain inherent to its design. It’s safer than January 2026, but not “safe” by enterprise standards. **How does it compare to Claude Code?** Claude Code is a session-based coding assistant with sandboxed execution and approval gates. OpenClaw is a persistent autonomous agent with full system access. Use Claude Code for production development; use OpenClaw (carefully) for personal automation that requires cross-app integration. **Why did the creator join OpenAI?** Peter Steinberger’s move to lead “personal AI agents” at OpenAI in February 2026 suggests the major platforms see OpenClaw’s architecture as the future direction. It also raises questions about long-term stewardship of the open-source project, which has transitioned to a foundation structure. **Is the “Linux of AI” comparison accurate?** Architecturally yes—the role of “agent OS” is analogous to Linux’s role as “hardware OS.” But security-wise, the comparison obscures critical differences. Linux was inert; OpenClaw is autonomous. The security model needs to evolve beyond what worked for traditional open source. **What’s the safest way to experiment?** Run NanoClaw instead for containerized isolation, or use OpenClaw only on dedicated hardware with no sensitive data. Never install skills without reviewing their permission requests. Consider managed platforms like Clawctl that handle hardening automatically. ## The Infrastructure Decision OpenClaw isn’t a toy, and it isn’t a tragedy. It’s an early version of the infrastructure we’ll need for autonomous AI—rough around the edges, powerful in the right hands, and genuinely dangerous in the wrong ones. The question “Linux or nightmare?” presents a false choice. Linux _was_ a nightmare for security teams in 1995—vulnerable, unproven, maintained by distributed volunteers. It became infrastructure because organizations invested in hardening it, and because the economic case was overwhelming. OpenClaw’s economic case is compelling: $5-50/month in API costs versus $20-200/month for Claude Code subscriptions, with full data sovereignty and no vendor lock-in. The security case is still being written. Whether it follows Linux’s trajectory or becomes a cautionary tale depends on whether the community can evolve security models as fast as the capabilities have evolved. For now, treat it like any powerful tool from an earlier era of computing: exciting, transformative, and absolutely not ready for production without significant guardrails. The future of open AI agents probably looks something like OpenClaw. Just don’t assume that future has arrived safely yet. ## Comments ## Related Articles [// Linux](/articles/rust-in-linux-kernel) ### [Rust in the Linux Kernel: One Year Later](/articles/rust-in-linux-kernel) [2026-01-22](/articles/rust-in-linux-kernel) [// ai](/articles/ai-in-india-2026-scared-leading-and-wasting-water) ### [AI in India 2026: Scared, Leading, and Wasting Water](/articles/ai-in-india-2026-scared-leading-and-wasting-water) [2026-05-16](/articles/ai-in-india-2026-scared-leading-and-wasting-water) [// ai](/articles/when-claude-became-my-relationship-coach-she-hated-ai) ### [When Claude Became My Relationship Coach (She Hated AI)](/articles/when-claude-became-my-relationship-coach-she-hated-ai) [2026-05-11](/articles/when-claude-became-my-relationship-coach-she-hated-ai) ### Never miss an update Join 50,000+ developers getting our weekly tech insights. [YOUR EMAIL ADDRESS]() Subscribe --- # What's New in ViteLand: February 2026 Recap > A comprehensive recap of all the exciting updates and developments in the Vite ecosystem for February 2026. // Tech 2026-03-02 [Jainil Prajapati](/authors/jainil-prajapati) 2026-03-02 - [Major Releases](#major-releases) - [Vite 6.0](#vite-60) - [Rolldown](#rolldown) - [Community Highlights](#community-highlights) - [Looking Ahead](#looking-ahead) The Vite ecosystem continues to evolve at a rapid pace. Here’s everything that happened in February 2026. ## Major Releases ### Vite 6.0 The latest major version of Vite brings significant performance improvements and new features: - **Faster cold starts** with improved dependency pre-bundling - **Better TypeScript support** with zero-config setup - **Enhanced plugin API** for more powerful integrations ### Rolldown The new Rust-based bundler continues to mature, with experimental support now available for production builds. ## Community Highlights The community has been incredibly active this month, with over 500 new plugins published to npm. ## Looking Ahead March promises even more exciting developments, including the official Vite+ announcement that took the community by storm. ## Comments ## Related Articles [// Tech](/articles/web-frameworks-benchmark) ### [VoidZero and npmx: Building Better Tools Together](/articles/web-frameworks-benchmark) [2026-03-03](/articles/web-frameworks-benchmark) [// AI](/articles/ai-models-2026) ### [Announcing Vite+ Alpha](/articles/ai-models-2026) [2026-03-13](/articles/ai-models-2026) [// ai](/articles/ai-in-india-2026-scared-leading-and-wasting-water) ### [AI in India 2026: Scared, Leading, and Wasting Water](/articles/ai-in-india-2026-scared-leading-and-wasting-water) [2026-05-16](/articles/ai-in-india-2026-scared-leading-and-wasting-water) ### Never miss an update Join 50,000+ developers getting our weekly tech insights. [YOUR EMAIL ADDRESS]() Subscribe --- # Qwen Just Became the Most Downloaded AI Model — Here's Why Nobody's Talking About It > Alibaba's Qwen hit 1 billion downloads. 8 of 10 top AI models on Hugging Face are Qwen. Then the guy who built it walked out. Here's what actually happened. // ai 2026-04-17 [Jainil Prajapati](/authors/jainil-prajapati) 2026-04-17 - [So what even is Qwen?](#so-what-even-is-qwen) - [The numbers are actually kind of insane](#the-numbers-are-actually-kind-of-insane) - [The race everyone is watching vs. the race that actually matters](#the-race-everyone-is-watching-vs-the-race-that-actually-matters) - [What Qwen actually shipped (while you were debating GPT-5)](#what-qwen-actually-shipped-while-you-were-debating-gpt-5) - [Why it matters that Qwen runs on a laptop in Ho Chi Minh City](#why-it-matters-that-qwen-runs-on-a-laptop-in-ho-chi-minh-city) - [Okay but let me actually steelman the counter-argument](#okay-but-let-me-actually-steelman-the-counter-argument) - [What the March crisis actually revealed](#what-the-march-crisis-actually-revealed) - [People are asking the right questions — just on the wrong model](#people-are-asking-the-right-questions--just-on-the-wrong-model) On March 4, 2026, a 32-year-old engineer named Lin Junyang posted five words on X: _“me stepping down. bye my beloved qwen.”_ That’s it. No thread. No long goodbye. Within hours, Alibaba’s stock dropped 4% in Hong Kong. An emergency all-hands was called. The CEO showed up. The post-training lead announced he was out too. The coding lead had already quietly left for Meta in January. Western AI Twitter noticed. Filed it under “Chinese tech drama.” Moved on. That was a mistake. --- ## So what even is Qwen? Qwen is Alibaba’s open-source AI model family — and as of 2026, the most downloaded on Earth with over 1 billion pulls on Hugging Face. It has spawned 200,000+ derivative models, surpassed Llama as the dominant fine-tuning base globally, and ships entirely under Apache 2.0 — meaning anyone can use it commercially, for free, no restrictions. --- ## The numbers are actually kind of insane Let me do the classic self-brag paragraph — except it’s not even my brag, it’s Alibaba’s. As of February 2026, **8 out of the 10 most downloaded text generation models on Hugging Face are from the Qwen family.** Not 3. Not 5. Eight. The most downloaded single model — Qwen2.5-7B-Instruct — has 13.3 million downloads. The second most downloaded is Qwen3-0.6B at 10.2 million. Llama-3.1-8B-Instruct shows up at number six. And all of this was built by a core team of roughly **100 people.** ByteDance’s foundational model team (Seed) has close to 2,000. Yuh. Seriously. A \~100-person team built the most-downloaded AI family on the planet, while operating with fewer compute resources than most of their competitors. Multiple Qwen insiders confirmed this to 36Kr. > **That’s not a product win. That’s one of the most remarkable resource efficiency stories in the history of software.** --- ## The race everyone is watching vs. the race that actually matters Here’s the thing about Western AI coverage. It’s obsessed with the capability race. Which model scores highest on MMLU. Which one passed the bar exam. Which one GPT-5 outperforms this week. These things matter. But there’s a second race happening simultaneously, and it’s arguably more consequential. **The infrastructure race.** Not “which model is smartest today” — but “which model gets embedded into the tools, fine-tunes, startups, and applications being built right now.” Because when a model becomes the default fine-tuning base, something structural happens. Every fine-tune inherits its architecture, its multilingual capabilities, its training biases, its default behaviors. That’s not a metaphor. That’s literally how fine-tuning works. 200,000 derivative models means 200,000 applications that **inherited something from Qwen.** And the Western AI narrative kept missing this — completely — because it was tracking benchmark leaderboards while Alibaba was quietly claiming the open-source infrastructure layer one Apache 2.0 release at a time. --- ## What Qwen actually shipped (while you were debating GPT-5) Okay this part matters for context. Alibaba launched Qwen in April 2023. Then opened it publicly after regulatory clearance that September. Then came Qwen2 (June 2024), Qwen2.5 (September 2024), Qwen3 (April 2025, trained on 36 trillion tokens across 119 languages), and then Qwen 3.5 in February 2026 — a 397B parameter model, natively multimodal, supporting 201 languages, Apache 2.0, benchmarking against frontier closed models. Then on March 30, 2026 — three weeks after the drama — **Qwen3.5-Omni dropped.** Real-time multimodal. 113-language speech recognition. State of the art on 215 audio/visual tasks. The same week, Qwen3.5-Max-Preview entered the LMArena Top 10 — surpassing GPT-5.4 and Claude Opus 4.5 in Expert Prompts. The Qwen app hit **203 million monthly active users in February 2026**, up from 31 million in January. It now sits third globally behind ChatGPT and ByteDance’s Doubao. I know this sounds like a press release. But the numbers are from Reuters and AICPB. The departures happened. The models are still shipping. **Both things are true at once.** --- ## Why it matters that Qwen runs on a laptop in Ho Chi Minh City This is the part that gets lost. The Western AI narrative is mostly written by people for whom Claude and GPT-5 are obvious defaults. API pricing is annoying, not prohibitive. But that’s not the global story. Businesses across Southeast Asia, the Middle East, North Africa, and Latin America are gravitating toward Chinese open-source models specifically because of accessibility. Free weights. No API dependency. Broad language support. Local deployment. The Qwen3-0.6B model — **600 million parameters** — runs on basically anything. Including a budget laptop. Including a tiny server in a market where OpenAI’s pricing structure is genuinely prohibitive. Chinese AI models’ share of total AI usage on OpenRouter hit nearly 30% by late 2025, up from 13% at the start of that year. That’s not benchmarks. That’s adoption. And **that’s where the real influence compounds.** The model running in Ho Chi Minh City isn’t GPT-5. It’s probably Qwen. \[INTERNAL LINK: What is fine-tuning and why does the base model matter?] --- ## Okay but let me actually steelman the counter-argument Not gonna lie — the “Qwen already won” framing is too clean. My first thought when I read the download numbers was: downloads aren’t deployments. A developer downloading Qwen to experiment is not the same as a hospital system running it in production. And that’s real. Practitioner consensus in 2026 still breaks roughly like this: **use Qwen for cutting-edge capabilities in experiments, but closed frontier models for production.** The reliability gap in high-stakes enterprise workflows is real. Benchmark numbers from the Qwen team’s own technical reports should be read with that context. Also — and this is the thing the “Qwen won the infrastructure war” take quietly slides past — **Apache 2.0 means anyone can fork it.** The 200,000 derivative models aren’t ideologically locked to Alibaba. If a better open base model shows up tomorrow, developers will move. So here’s what I actually think: The “danger” of Qwen isn’t that it replaces GPT-5 in enterprise contracts. It won’t. Not this year. The danger is **structural and slow.** It’s that the default substrate of global open-source AI development is increasingly built on Alibaba’s architectural choices, Alibaba’s training data, Alibaba’s multilingual biases. That compounds outward — invisibly, across 200,000 applications — in ways that are very hard to reverse. That’s a different kind of influence. And it doesn’t show up on a benchmark leaderboard. --- ## What the March crisis actually revealed Here’s the thing about the Lin Junyang situation that most coverage missed. The core Qwen team was \~100 people. Lin had been pushing since 2025 to keep the team **vertically integrated** — pre-training, post-training, language, multimodal, code — all working together, in tight sync. Alibaba corporate disagreed. They wanted to restructure into horizontally specialized units. Split the team by function. Merge the pieces with other Tongyi Lab units. Scale it up, enterprise-style. Lin walked out of a heated meeting and submitted his resignation the next day. This is not really a story about one engineer’s ego. This is a story about **what made Qwen work in the first place.** A small, tight team with fewer resources than competitors — moving fast, staying integrated, shipping five major model generations in three years. That’s the culture that produced the most-downloaded AI model family on Earth. Alibaba looked at that success and decided: we need to turn this into a proper organization. And the guy who built it said: no thanks. Whether you think that’s the right call by Alibaba or not — that’s a genuinely hard question — the fear in the open-source community is obvious. Not that Qwen stops being capable. But that it stops being **genuinely open.** That the commercial pressures win. That the Qwen App’s DAU metrics start driving decisions that used to be made by researchers chasing the frontier. You don’t convene an emergency all-hands with the CEO over a product that doesn’t matter. That’s the tell. --- ## People are asking the right questions — just on the wrong model **Is Qwen open-source?** Yes. Fully. Apache 2.0. Every weight. Commercial use included. **Can Qwen run locally?** Absolutely. The 0.6B model runs on a CPU. The 7B runs on a mid-range consumer GPU. The 27B fits on a single 32GB VRAM card. Ollama, llama.cpp, LM Studio — all supported. **Is Qwen better than ChatGPT?** Depends entirely on what you’re doing. For multilingual tasks, math, and coding? Qwen 3.5-27B wins or ties on most benchmarks. For high-stakes enterprise production workloads with reliability requirements? Closed frontier models still have an edge. **What happened to Qwen’s lead developer?** Lin Junyang resigned in March 2026 after disagreeing with Alibaba’s plan to restructure the team. He hasn’t announced his next move. The AI world is watching. **What is the most downloaded AI model in 2026?** Qwen2.5-7B-Instruct, with 13.3 million downloads — and it’s one of eight Qwen models in the top ten. --- Anyway. The question worth sitting with isn’t whether Qwen beats GPT-5. It’s whether the tools being built right now — the healthcare apps, the legal fine-tunes, the startups in markets you don’t write about — are running on architecture that came out of a 100-person team in Hangzhou. Because that’s already happened. Whether it keeps happening after the restructuring is the actual story of 2026. PS: If you’re building anything that touches open-source LLMs and you haven’t seriously evaluated Qwen 3.5 yet — genuinely curious what’s kept you away. Drop it in the comments. ## Comments ## Related Articles [// ai](/articles/ai-in-india-2026-scared-leading-and-wasting-water) ### [AI in India 2026: Scared, Leading, and Wasting Water](/articles/ai-in-india-2026-scared-leading-and-wasting-water) [2026-05-16](/articles/ai-in-india-2026-scared-leading-and-wasting-water) [// ai](/articles/when-claude-became-my-relationship-coach-she-hated-ai) ### [When Claude Became My Relationship Coach (She Hated AI)](/articles/when-claude-became-my-relationship-coach-she-hated-ai) [2026-05-11](/articles/when-claude-became-my-relationship-coach-she-hated-ai) [// ai](/articles/tokens-are-getting-cheaper-so-why-does-ai-feel-more-expensive) ### [Tokens Are Getting Cheaper. So Why Does AI Feel More Expensive?](/articles/tokens-are-getting-cheaper-so-why-does-ai-feel-more-expensive) [2026-05-08](/articles/tokens-are-getting-cheaper-so-why-does-ai-feel-more-expensive) ### Never miss an update Join 50,000+ developers getting our weekly tech insights. [YOUR EMAIL ADDRESS]() Subscribe --- # Rust in the Linux Kernel: One Year Later > Reflecting on one year of Rust programming language integration into the Linux kernel development. // Linux 2026-01-22 [Jainil Prajapati](/authors/jainil-prajapati) 2026-01-22 - [The Beginning](#the-beginning) - [Progress Made](#progress-made) - [Challenges](#challenges) - [Conclusion](#conclusion) One year ago, the Linux kernel community made history by accepting the first Rust code into the mainline kernel. Now, we take stock of what’s been accomplished and what lies ahead. ## The Beginning The decision to introduce Rust into the kernel was met with both excitement and skepticism. Would the memory safety guarantees translate into real-world benefits? Could Rust coexist peacefully with decades of C code? ## Progress Made Over the past year: - **12 new Rust drivers** merged into mainline - **4 subsystem maintainers** actively reviewing Rust patches - **Significant bug reduction** in Rust-written components - **Improved tooling** for kernel development ## Challenges Working with Rust in a kernel environment presents unique challenges: 1. **No standard library** - Everything must be written for `no_std` 1. **Embedded development** - No heap allocation by default 1. **Abstraction boundaries** - Bridging Rust and C safely ## Conclusion The experiment has been a success. Rust is now a first-class citizen in the Linux kernel, with growing adoption across multiple subsystems. ## Comments ## Related Articles [// Linux](/articles/open-claw-the-linux-of-ai-agents-or-a-security-nightmare) ### [OpenClaw: The Linux of AI Agents or a Security Nightmare?](/articles/open-claw-the-linux-of-ai-agents-or-a-security-nightmare) [2026-03-22](/articles/open-claw-the-linux-of-ai-agents-or-a-security-nightmare) [// ai](/articles/ai-in-india-2026-scared-leading-and-wasting-water) ### [AI in India 2026: Scared, Leading, and Wasting Water](/articles/ai-in-india-2026-scared-leading-and-wasting-water) [2026-05-16](/articles/ai-in-india-2026-scared-leading-and-wasting-water) [// ai](/articles/when-claude-became-my-relationship-coach-she-hated-ai) ### [When Claude Became My Relationship Coach (She Hated AI)](/articles/when-claude-became-my-relationship-coach-she-hated-ai) [2026-05-11](/articles/when-claude-became-my-relationship-coach-she-hated-ai) ### Never miss an update Join 50,000+ developers getting our weekly tech insights. [YOUR EMAIL ADDRESS]() Subscribe --- # Tech Industry Hiring Stabilizes After Tumultuous Year > Analysis of the tech hiring landscape in early 2026 and what it means for developers and companies. // Tech 2026-02-28 [Jainil Prajapati](/authors/jainil-prajapati) 2026-02-28 - [The Rollercoaster](#the-rollercoaster) - [Current State](#current-state) - [What This Means for Developers](#what-this-means-for-developers) - [Opportunities](#opportunities) - [Challenges](#challenges) - [Looking Forward](#looking-forward) After years of layoffs and uncertainty, the tech industry is finally seeing stabilization in hiring patterns. ## The Rollercoaster The past few years have been tumultuous for tech workers: - 2023 saw massive layoffs across the industry - 2024 brought continued uncertainty - 2025 marked the beginning of recovery ## Current State Today, hiring has returned to more sustainable levels: - **20% increase** in job postings compared to 2025 - **Focus on quality over quantity** - **Higher emphasis on AI/ML skills** - **Remote work** remains the preferred mode ## What This Means for Developers The stabilization brings both opportunities and challenges: ### Opportunities - More job security - Competitive compensation - Better work-life balance ### Challenges - Higher expectations for productivity - Need for continuous skill development - Competition for premium positions ## Looking Forward The industry appears to be entering a new phase of sustainable growth, with companies focusing on building efficient, productive teams rather than chasing rapid expansion. ## Comments ## Related Articles [// ai](/articles/ai-in-india-2026-scared-leading-and-wasting-water) ### [AI in India 2026: Scared, Leading, and Wasting Water](/articles/ai-in-india-2026-scared-leading-and-wasting-water) [2026-05-16](/articles/ai-in-india-2026-scared-leading-and-wasting-water) [// ai](/articles/when-claude-became-my-relationship-coach-she-hated-ai) ### [When Claude Became My Relationship Coach (She Hated AI)](/articles/when-claude-became-my-relationship-coach-she-hated-ai) [2026-05-11](/articles/when-claude-became-my-relationship-coach-she-hated-ai) [// Tech](/articles/apple-vision-light-the-ar-glasses-we-ve-been-waiting-for) ### [Apple Vision Light: The AR Glasses We've Been Waiting For?](/articles/apple-vision-light-the-ar-glasses-we-ve-been-waiting-for) [2026-03-20](/articles/apple-vision-light-the-ar-glasses-we-ve-been-waiting-for) ### Never miss an update Join 50,000+ developers getting our weekly tech insights. [YOUR EMAIL ADDRESS]() Subscribe --- # Tokens Are Cheap. Thinking Isn't. > I put a daily limit on AI prompts. Not because I can't afford the tokens, but because my brain can't afford the attention. Here's why limiting daily AI prompts protects clarity. // ai 2026-04-24 [Jainil Prajapati](/authors/jainil-prajapati) 2026-04-24 - [Why Limiting Daily AI Prompts Protects Your Clarity](#why-limiting-daily-ai-prompts-protects-your-clarity) - [The 4-hour ceiling is real](#the-4-hour-ceiling-is-real) - [The anxiety loop](#the-anxiety-loop) - [So I put a number on it](#so-i-put-a-number-on-it) - [FAQ](#faq) - My first prompt of the day is always great. By prompt twelve, I’m typing “fix this” and getting mad when the output sucks. That’s when I realized the cost isn’t financial anymore. It’s cognitive. ## Why Limiting Daily AI Prompts Protects Your Clarity Limiting daily AI prompts protects your clarity because every prompt costs attention, not just tokens. When you cap your messages, you force yourself to think before you send, which breaks the anxiety loop and restores decision-making quality. Here’s the thing. We talk about token prices like they’re the only cost. GPT-3 was $60 per million tokens in 2021. By early 2026, equivalent performance cost $0.06. That’s a 1,000x drop. Enterprise AI spending still grew 320% in the same window because everyone is using it more. But nobody’s talking about the other invoice. The one your brain pays. Every prompt is a micro-decision. Every follow-up is a judgment call. Every “hmm, that’s not quite right” is you holding context, evaluating output, and deciding what matters. The machine generates; the human judges. And judging at speed is one of the most depleting forms of cognitive work. > AI compresses the effort required for production, but it doesn’t compress the cognitive cost of judgment. I read that somewhere and it broke my brain a little. ## The 4-hour ceiling is real There’s a study from BCG and Harvard Business Review, published March 2026. They coined the term “AI brain fry” — mental fatigue from excessive oversight of AI tools. Turns out people managing multiple AI agents expend 14% more mental effort and report 39% higher error rates. Another engineer put it perfectly: after 4-5 hour sessions, prompt quality degrades before you even notice you’re tired. Yuh. That tracks. My job used to be: think, create, ship. Now it’s: prompt, wait, evaluate, fix, re-prompt, repeat. I became a reviewer. A quality inspector on an assembly line that never stops. And honestly? AI-generated work requires _more_ careful review than the human version. Which is stupid if you ask me. ## The anxiety loop If you’re going out with your girlfriend, you feel like you should have an AI agent running in the background. If you’re brushing your teeth, maybe you should have already prompted something. It slowly starts to feel like there’s no other way to operate. I know this sounds stupid as I’m writing it. But that’s exactly what happens. The tool that was supposed to save time starts consuming the in-between moments. The gaps where thinking used to happen. And when the output gets worse because you’re tired, you send _more_ prompts to fix it. Which makes you more tired. It’s a loop. And loops are heavy. ## So I put a number on it Not because I can’t afford it. Luckily, I can. But because my brain can’t. I don’t know what the exact number is yet. Maybe five messages. Maybe ten. But the idea is simple: if there’s a limit, every message has to count. I have to think before I send it. This isn’t some productivity hack. It’s not about saving money. It’s about protecting the only thing that doesn’t scale: your ability to know what good looks like without needing a second opinion from a machine. When you know you only have five prompts, you don’t waste one on “make this better.” You sit with the problem first. You get specific. You actually think. And weirdly? The output improves. Not because the model got smarter, but because my prompts did. ## FAQ **What is AI brain fry?** It’s acute cognitive fatigue from managing AI tools beyond your mental capacity. BCG researchers named it in March 2026. Symptoms include mental static, decision fatigue, and that “dozen browser tabs open in my head” feeling. **Why do I feel more tired using AI?** Because you’re doing more evaluative work, not less. Creating gives you flow states. Reviewing gives you decision fatigue. AI shifted your job from maker to judge. **How many AI prompts per day is too many?** There’s no universal number, but quality degrades after sustained use. Some practitioners call the “4-hour ceiling” the hard limit for productive AI-assisted work. For me, it’s about counting messages, not hours. **Is AI actually making us more productive?** Yes, but with a paradox. Individual tasks get faster while total workload expands. AI lowers the cost of production but raises the cost of coordination, review, and decision-making. Those costs fall entirely on you. **How do I avoid AI fatigue?** Set hard boundaries. Cap your daily prompts. Force no-AI blocks for deep work. Stop treating “always on” as a virtue. Your value isn’t your ability to prompt — it’s your ability to discern. #### Anyway. Stop waiting. Start thinking. Chalo, bye! PS: If you catch me sending “fix this” at 11pm, please tell me to log off. ## Comments ## Related Articles [// ai](/articles/ai-in-india-2026-scared-leading-and-wasting-water) ### [AI in India 2026: Scared, Leading, and Wasting Water](/articles/ai-in-india-2026-scared-leading-and-wasting-water) [2026-05-16](/articles/ai-in-india-2026-scared-leading-and-wasting-water) [// ai](/articles/when-claude-became-my-relationship-coach-she-hated-ai) ### [When Claude Became My Relationship Coach (She Hated AI)](/articles/when-claude-became-my-relationship-coach-she-hated-ai) [2026-05-11](/articles/when-claude-became-my-relationship-coach-she-hated-ai) [// ai](/articles/tokens-are-getting-cheaper-so-why-does-ai-feel-more-expensive) ### [Tokens Are Getting Cheaper. So Why Does AI Feel More Expensive?](/articles/tokens-are-getting-cheaper-so-why-does-ai-feel-more-expensive) [2026-05-08](/articles/tokens-are-getting-cheaper-so-why-does-ai-feel-more-expensive) ### Never miss an update Join 50,000+ developers getting our weekly tech insights. [YOUR EMAIL ADDRESS]() Subscribe --- # Tokens Are Getting Cheaper. So Why Does AI Feel More Expensive? > Prime's right that the AI economy is shifting — but the real story isn't about money. It's about GPUs, and nobody has enough of them. Let's break this down. // ai 2026-05-08 [Jainil Prajapati](/authors/jainil-prajapati) 2026-05-08 - [Okay so what’s actually happening with AI pricing right now?](#okay-so-whats-actually-happening-with-ai-pricing-right-now) - [Here’s what actually happened — the timeline](#heres-what-actually-happened--the-timeline) - [The Cursor moment (the real start)](#the-cursor-moment-the-real-start) - [Anthropic’s compute rationing experiments](#anthropics-compute-rationing-experiments) - [The Claude Code drama](#the-claude-code-drama) - [GitHub pausing signups](#github-pausing-signups) - [The compute economics that people keep skipping over](#the-compute-economics-that-people-keep-skipping-over) - [The pre-training vs post-training thing matters here](#the-pre-training-vs-post-training-thing-matters-here) - [The Google take in Prime’s video is just wrong, bruh](#the-google-take-in-primes-video-is-just-wrong-bruh) - [Okay but tokens are getting cheaper, right?](#okay-but-tokens-are-getting-cheaper-right) - [The real story nobody’s telling](#the-real-story-nobodys-telling) - [FAQ](#faq) I had a single message running in Copilot for over two hours. One message. One. It was trying to crack a cryptography challenge and just… kept going. And the whole time, my Copilot usage meter was sitting at 0.4%. So when people started panicking about GitHub pausing signups or Anthropic messing with the $20 Claude plan — I had an actual answer. Not vibes. Not a hot take. Let me explain what’s actually going on here, because the discourse has been genuinely bad. --- ## Okay so what’s actually happening with AI pricing right now? The short version: **Anthropic, GitHub, and Cursor aren’t raising prices to make more money. They’re running out of GPUs.** These companies don’t have enough compute to keep subsidizing the level of usage they’ve been giving away. And as agentic workflows explode — where a single “message” can run for hours, use $100+ of inference, and spawn parallel threads — the old flat-rate subscription math completely falls apart. This isn’t the end of cheap AI. It’s the end of unlimited AI. There’s a difference. --- ## Here’s what actually happened — the timeline Prime made a video about the “cracks showing in the AI economy.” He’s not wrong that something is shifting. But I think he missed some important details, especially around the economics of it. Let me do the context first. #### The Cursor moment (the real start) This actually goes back further than the recent drama. Cursor was one of the first to feel it. They were pricing by number of messages. Some messages cost a few cents to run. Others cost several hundred dollars. And Cursor has to pay the labs — Anthropic, OpenAI — in actual cash for each request. They couldn’t eat the variance. A heavy user could consume $3,000 of inference on a $200 plan. So they switched to usage-based pricing — your cost reflects what it actually costs to run your requests. That was the writing on the wall. That was 2024. #### Anthropic’s compute rationing experiments In March this year, Anthropic ran something subtle. For two weeks, they doubled usage outside of peak hours. Sounds generous, right? It wasn’t altruism. It was them trying to push power users off-peak so their GPUs would be available during business hours when enterprise customers needed them. It didn’t work fast enough. A week and a half later, they quietly announced that between 5am and 11am PT, you’d burn through your session limits faster than before. They tried the carrot. Then used the stick. #### The Claude Code drama April 21st, Anthropic’s pricing page briefly showed Claude Code was only available on the $100 Max plan — not the $20 Pro plan. Twitter lost it. Sam Altman posted something about being drunk. Memes happened. Anthropic said it was a test on 2% of new signups and they reverted it. Here’s the thing people missed: **this wasn’t them trying to push you from $20 to $100.** They don’t actually care that much about moving you up a tier. What they care about is enterprise revenue — which is where Anthropic is now at $30B ARR, ahead of OpenAI. The $20 plan change was them trying to slow down the compute firehose from low-tier users. Same energy as what GitHub did a week later. #### GitHub pausing signups April 20th. GitHub pauses new Copilot Pro and Pro+ signups entirely. Copilot VP Joe Binder wrote: _“It’s now common for a handful of requests to incur costs that exceed the plan price.”_ That’s the sentence. That’s the whole thing. Agentic coding sessions — where you give Claude or GPT a task and it runs for hours, spawns subagents, debugs automatically — were never what the original Copilot pricing was built for. The old model was autocomplete. Short, stateless suggestions. Now people are running multi-hour parallel agent sessions on a $40/month plan. > **You don’t pause signups because you want more money. You pause signups because you don’t have capacity.** Microsoft isn’t compute-poor because they’re bad at business. They’re compute-constrained because everyone is. And the GPUs that $40/month developers are using? Microsoft needs those for the Fortune 500 companies paying actual enterprise rates. --- ## The compute economics that people keep skipping over Let me be real about the numbers here. A $200/month Claude Max subscription was being used for somewhere between $1,000 and $5,000 of compute. According to people tracking this closely, Anthropic was subsidizing at 20x or more. That’s not unsustainable because of the subscription revenue gap alone. **It’s unsustainable because of electricity.** Running high-end GPUs at full load is genuinely expensive. I have a 5090 doing some training tasks. It alone bumped my electric bill by roughly $1,000. That’s one consumer card. Anthropic is running server farms. Estimates put the raw compute cost to Anthropic at somewhere around 15–20% of what you pay on the API. So they’re not just losing the revenue gap — there are users in the $200/month tier who are legitimately costing Anthropic money on electricity alone, before you account for the GPU depreciation, the training costs, the engineers, any of it. #### The pre-training vs post-training thing matters here Prime made a point about Opus 4.7 being less impressive and thus potentially losing money on the model drop. I understand where that’s coming from but I think it misses something important. Not every model release involves a full new pre-training run. Pre-training is the big, insanely expensive thing. That’s where you take terabytes of data and compress it into the model’s weights. For a frontier model? We’re talking hundreds of millions to potentially billions of dollars. Post-training — the fine-tuning, the RLHF, the RLVR — is how you shape the model’s behavior. It’s gotten incredibly powerful (it’s why agentic coding performance has jumped so much recently). And it’s often significantly cheaper than pre-training. When you see a model drop that feels like “the same but a bit better,” that’s probably post-training work. When you see a model that feels fundamentally different — like going from GPT-5.3 to 5.5 — that’s likely a new pre-training. Opus 4.5 was probably new pre-training. That’s why it felt so different and got way cheaper to run than previous Opus models. Opus 4.6 and 4.7? More likely post-training iterations. Less expensive. Not “failures” — just a different cost structure. So the claim that they’re losing money on model drops because newer models don’t get as much adoption doesn’t really hold in this framing. The economics per model depend heavily on what type of training was done. --- ## The Google take in Prime’s video is just wrong, bruh Prime frames Google as the company that doesn’t have this problem because they make money elsewhere and can keep subsidizing. I get why it looks that way from the outside. But this is actually backwards. **Google was subsidizing harder than anyone.** Anti-gravity — sorry, Google’s AI suite — had Opus 4.5 included in it. I personally knew people who were on Google’s subscription purely for the subsidized Opus access. And that exploded so badly that Google was the **first** company to start aggressively restricting usage. People building plugins to track their usage? Banned. People linking Anti-gravity to open code tools? Banned. Quick restrictions, hard limits. Google didn’t avoid this problem because they have money. They hit it first and had to walk it back faster and more aggressively than anyone else. The reason you don’t think of Google as part of this story? Their models have been mediocre enough that fewer developers noticed or cared. It doesn’t make the news when Google tightens limits because nobody’s super emotionally attached to Gemini’s free tier. But Google is actually the most extreme example of exactly what Prime’s talking about. They’re just underreported because of the model quality issue. --- ## Okay but tokens are getting cheaper, right? Here’s where my actual opinion comes in. **Yes. And also no. And the nuance is the whole point.** Token prices per million are going up at the frontier. GPT-5.5 is 2x the token cost of 5.4. Fact. But something really interesting is happening underneath that number. The models are getting more efficient. 5.5 uses significantly fewer tokens per task than 5.4 did — especially for longer prompts where it uses 19–34% fewer completion tokens. The OpenRouter team did the actual analysis: switching from 5.4 to 5.5 raised real-world costs by 49–92% depending on your prompt length. Not 2x. The efficiency partially offsets the price hike. And then look at what the mid-tier does. GPT-5.5 medium is just as smart as 5.4 was at peak. Same benchmark scores. But costs less than half as much to run. If you were happy with 5.4’s intelligence, you’re paying substantially less for the same capability on the next model cycle. That’s what actually matters. **At any given level of intelligence, the cost is dropping consistently.** The frontier gets more expensive as more compute gets poured in. But last-gen frontier becomes this-gen mid-tier at a fraction of the price. > The cost per unit of intelligence is going down. The cost of access to the absolute bleeding edge is going up. Both are true. The problem is that the access restrictions feel worse because they’re visible and immediate, while the efficiency gains happen quietly underneath. --- ## The real story nobody’s telling Here’s the thing Prime is gesturing at but doesn’t quite land on: **This isn’t the end of the subsidy economy. It’s the start of compute restrictions actually mattering.** The Anthropic $20 plan change and the GitHub Copilot signup pause are the same event. They both have the same cause. Neither company has enough Nvidia GPUs in their server farms to serve their enterprise customers and keep giving away compute at flat-rate consumer prices. The bottleneck isn’t money. Microsoft has money. Anthropic has $18 billion in funding. The bottleneck is physical hardware that takes 12–24 months to order, build infrastructure for, and deploy. OpenAI had this same problem in 2022. Sam literally paused ChatGPT Plus signups back then for the same reason. They just bought more compute aggressively and now they don’t have this problem. Not because they have more money — because they have more chips. Chips can’t be made fast enough. That’s the real story. When you frame it as “these companies are being greedy,” you end up in conspiracy territory that just isn’t accurate. The Copilot 7.5x message multiplier for GPT-5.5 isn’t based on what the API costs. It’s based on how much compute Microsoft has provisioned for that model cluster and how much of it they’re willing to let $40/month users consume when enterprise deals are competing for the same GPUs. And by the way — Uber spent their entire year’s AI budget in four months. They told every employee to use AI maxally, then acted surprised. That’s API usage at full enterprise rates. Not $200/month subs. The average engineer at a company like Uber doing heavy AI usage is probably doing similar inference volume to what I do on my $200/month plan. They pay $2,000+ for it. That gap is the whole game. --- ## FAQ **Why did GitHub pause Copilot signups?** Because agentic coding sessions — where you hand AI a task and it runs autonomously for hours — now routinely cost more compute than the monthly plan price in a single session. GitHub VP Joe Binder literally said “it’s now common for a handful of requests to incur costs that exceed the plan price.” This isn’t revenue collection. It’s compute triage. **Is the all-you-can-eat AI era actually over?** The unlimited era is over, yes. But “all you can eat” was always a temporary subsidization play to build market share. The smarter reframe: at any level of intelligence you actually need, the costs are dropping. You just can’t run a $5,000 agentic session for $200/month anymore. **Are AI tokens genuinely getting cheaper?** Per unit of intelligence, yes. Per token at the frontier, no — costs are going up as models get more capable. The trick is that the same level of intelligence that cost $X six months ago now costs less. The frontier keeps moving. **Why did Anthropic remove Claude Code from the $20 plan?** They briefly tested it on 2% of new signups, it caused a Twitter meltdown, Sam Altman dunked on them while apparently half in the bag, and they reverted it. The actual goal wasn’t to push people to $100. It was to slow down compute consumption from the lowest-tier users who weren’t their target customer anyway. **Is the AI bubble about to burst?** No. But the “unlimited compute as a marketing play” era is definitely ending. The companies who have more GPUs will win. The ones that don’t will ration. That’s not a bubble pop — that’s a hardware supply problem. --- The real question isn’t “are tokens getting cheaper.” It’s “cheaper than what, for whom, and measured how.” Per task? Getting cheaper. Per unit of intelligence? Getting cheaper. Per month on a flat consumer sub while running 24/7 agentic workloads? That math was always fake, and now they’re fixing it. That’s all. PS: That one Copilot message eventually finished after 2h15m. And my weekly usage is still at like 0.4%. So. Yeah. Chalo, bye! ## Comments ## Related Articles [// ai](/articles/ai-in-india-2026-scared-leading-and-wasting-water) ### [AI in India 2026: Scared, Leading, and Wasting Water](/articles/ai-in-india-2026-scared-leading-and-wasting-water) [2026-05-16](/articles/ai-in-india-2026-scared-leading-and-wasting-water) [// ai](/articles/when-claude-became-my-relationship-coach-she-hated-ai) ### [When Claude Became My Relationship Coach (She Hated AI)](/articles/when-claude-became-my-relationship-coach-she-hated-ai) [2026-05-11](/articles/when-claude-became-my-relationship-coach-she-hated-ai) [// ai](/articles/tokens-are-cheap-thinking-isnt) ### [Tokens Are Cheap. Thinking Isn't.](/articles/tokens-are-cheap-thinking-isnt) [2026-04-24](/articles/tokens-are-cheap-thinking-isnt) ### Never miss an update Join 50,000+ developers getting our weekly tech insights. [YOUR EMAIL ADDRESS]() Subscribe --- # VoidZero and npmx: Building Better Tools Together > Exploring the collaboration between VoidZero and npmx to create better JavaScript development tools. // Tech 2026-03-03 [Jainil Prajapati](/authors/jainil-prajapati) 2026-03-03 - [VoidZero’s Vision](#voidzeros-vision) - [npmx: The Modern Package Manager](#npmx-the-modern-package-manager) - [Why This Matters](#why-this-matters) - [The Future](#the-future) The JavaScript tooling ecosystem is consolidating, and two recent announcements highlight this trend: VoidZero’s new unified toolchain and the npmx package manager. ## VoidZero’s Vision VoidZero has been working on a next-generation JavaScript toolchain that aims to unify: - **Bundling** - Fast, efficient builds - **Minification** - Optimized output - **Linting** - Code quality tools - **Testing** - Comprehensive coverage ## npmx: The Modern Package Manager Building on lessons from npm, yarn, and pnpm, npmx brings: - **Faster installs** through intelligent caching - **Better deduplication** of dependencies - **Improved security** scanning - **Seamless migration** from existing package managers ## Why This Matters The fragmentation of the JavaScript tooling landscape has been a pain point for developers for years. These new tools promise a more cohesive, efficient development experience. ## The Future As these tools mature, we can expect: 1. Faster development workflows 1. Smaller bundle sizes 1. Better developer experience 1. Reduced configuration overhead ## Comments ## Related Articles [// Tech](/articles/quantum-computing-breakthrough) ### [What's New in ViteLand: February 2026 Recap](/articles/quantum-computing-breakthrough) [2026-03-02](/articles/quantum-computing-breakthrough) [// AI](/articles/ai-models-2026) ### [Announcing Vite+ Alpha](/articles/ai-models-2026) [2026-03-13](/articles/ai-models-2026) [// ai](/articles/ai-in-india-2026-scared-leading-and-wasting-water) ### [AI in India 2026: Scared, Leading, and Wasting Water](/articles/ai-in-india-2026-scared-leading-and-wasting-water) [2026-05-16](/articles/ai-in-india-2026-scared-leading-and-wasting-water) ### Never miss an update Join 50,000+ developers getting our weekly tech insights. [YOUR EMAIL ADDRESS]() Subscribe --- # When Claude Became My Relationship Coach (She Hated AI) > I deleted 1,080+ messages on Claude's advice. Then she found out I was using AI to talk to her. Here's where I drew the line — and where I should've started. // ai 2026-05-11 [Jainil Prajapati](/authors/jainil-prajapati) 2026-05-11 - [Okay, so here’s what actually happened](#okay-so-heres-what-actually-happened) - [So wait — is using AI for this actually okay?](#so-wait--is-using-ai-for-this-actually-okay) - [The part that nobody talks about](#the-part-that-nobody-talks-about) - [And she sniffed it out anyway](#and-she-sniffed-it-out-anyway) - [Where AI is actually useful in relationships (and where it’s not)](#where-ai-is-actually-useful-in-relationships-and-where-its-not) - [The real question she was asking](#the-real-question-she-was-asking) - [FAQ](#faq) I deleted more than 1,080 messages because an AI told me not to send them. And honestly? A part of me is still not sure that was a bad call. --- ## Okay, so here’s what actually happened I met this girl on Snapchat. We clicked immediately. Like, actually clicked — not the weird forced “haha same” kind of clicking. Real stuff. But here’s my problem with myself. **I care too much. Too fast.** And when I care too much, I go full clingy without even realizing it. She noticed. She said something. And I felt terrible about it — because the caring was real, but the execution was a disaster. We eventually worked things out and became friends again. And somewhere in that process, I started using Claude. Not as a ghostwriter. Not to fake things. I started using it as a check system. Like — before I’d send her a long paragraph about how I felt, I’d paste it into Claude first. “Is this too much? Should I send this?” If Claude said it was too heavy for where things were at, I’d delete it. Over and over. That became the whole system. More than 1,080 messages gone. And it actually worked. I was showing up less clingy. The conversations were better. She wasn’t pulling back. In my head, Claude was doing me a favour. Then I confessed. I told her I’d been asking AI what to share and how much to share. Every time. She was disappointed. --- ## So wait — is using AI for this actually okay? > Using AI for dating advice is useful when it helps you express feelings you already have — but dangerous when it starts replacing your own emotional judgment entirely. The line is between AI as a thinking partner and AI as your emotional brain. Once you can’t function without it, something’s gone wrong. That’s the core of it. And most people talking about this are missing that distinction completely. --- ## The part that nobody talks about Here’s what the “AI in romance = cheating” crowd gets wrong. They assume the problem is the AI. But the AI didn’t create clingy behavior. The anxiety was already there. The fear of losing her was already there. Claude just stepped in as a system to regulate something that was already dysregulated. Is that bad? Honestly — partially, no. There’s a study from the University of Kent (February 2026, nearly 4,000 participants) that found people who use AI for personal messages are seen as **less caring, less authentic, less trustworthy.** And that’s real. That tracks. But that’s a _perception_ problem. Not always a reality problem. For someone who genuinely struggles to know when they’re being too much — and I know I am — Claude gave me the thing I didn’t have: a pause. A second opinion. A “hey, maybe don’t.” The issue wasn’t using Claude. **The issue was that I couldn’t do this without Claude.** --- ## And she sniffed it out anyway Here’s the thing about AI-generated or AI-filtered messages. They’re clean. Too clean. Almost half of Gen Z already uses AI for dating advice (Match survey data), so people are starting to recognize the pattern. Not always consciously. But something feels slightly off. She didn’t notice while it was happening. But she knew something was weird. And when I told her, it all clicked. The disappointment wasn’t “you used a tool.” The disappointment was: **“You didn’t trust yourself with me.”** And I think that’s the more painful truth. --- ## Where AI is actually useful in relationships (and where it’s not) Let me be real about this. **Where AI is genuinely good:** - Understanding your own patterns. Asking “why do I always do this in relationships” — Claude is surprisingly thoughtful here. - Getting a gut check before a hard conversation. Not “write this for me” but “does this land the way I think it does?” - Processing a confusing situation when no one else is available. It’s a mirror, not an answer. **Where AI will genuinely mess you up:** - Copy-pasting responses. She’ll know. They always know. - Using it so much that you stop trusting your own instincts about what to say. - Letting it become the filter for every single thing — until your own voice disappears. > The problem isn’t asking AI what to say. The problem is forgetting how to decide for yourself. A Stanford researcher, Myra Cheng, found that LLMs have higher sycophancy rates than humans — meaning they’ll often just tell you what you want to hear. Which is the last thing you need when you’re already anxious about a relationship and looking for permission. \[INTERNAL LINK: AI sycophancy problem — when chatbots agree with everything] --- ## The real question she was asking When she said she was disappointed, she wasn’t saying “you shouldn’t use AI.” She was saying: “If Claude was controlling what you shared with me — then who was I actually talking to?” That’s a fair question. I don’t have a great answer. Because the feelings were mine. The care was mine. The context Claude had was everything I fed it. But somewhere in 1,080 deleted messages, a version of me that was messier and more honest and maybe more real — never made it through. Which is stupid if you ask me. But also kind of understandable. --- ## FAQ **Is it wrong to use AI to text your crush?** Not inherently. Using AI to help you communicate feelings you actually have isn’t fake. Using AI to generate feelings or personas you don’t actually have — that’s where it gets dishonest. **Can someone tell if you used AI to write your messages?** Often, yes. AI-filtered messages tend to be too clean, too coherent, too measured. If you usually type “u” and suddenly you’re writing full paragraphs with perfect commas — she’s going to notice something is up. **Should I use AI for relationship advice?** As a starting point, sure. It’s weirdly good at identifying patterns and helping you frame thoughts. But don’t take its side blindly — studies show it agrees with you more than it should. Always run it past real people too. **Is using AI for dating deceptive?** Using it as a ghostwriter = kinda yes. Using it as a thinking tool, then showing up with your own words = not really. The difference is whether the final message sounds and feels like you. --- Not gonna lie. I still use Claude sometimes. But now when I’m about to send something, I ask it differently. Not “should I send this” — but “what am I actually trying to say here?” That’s me using it as a mirror. Not me handing it the wheel. **PS**: If you’re doing the 1,080 deleted messages thing — maybe also try just talking to your actual friends lol. They’re less objective but they know you better. If the girl reading this somehow finds this post — I’m sorry for the filtered version. The unfiltered one cares for you more than Claude knew how to say. **PPS:** I’m not saying don’t use AI. I’m saying know what you’re actually asking it to do. Using it to understand yourself? Good. Using it to control how much of yourself someone else gets to see? That’s where it gets complicated. That’s where I got it wrong. ## Comments ## Related Articles [// ai](/articles/ai-in-india-2026-scared-leading-and-wasting-water) ### [AI in India 2026: Scared, Leading, and Wasting Water](/articles/ai-in-india-2026-scared-leading-and-wasting-water) [2026-05-16](/articles/ai-in-india-2026-scared-leading-and-wasting-water) [// ai](/articles/tokens-are-getting-cheaper-so-why-does-ai-feel-more-expensive) ### [Tokens Are Getting Cheaper. So Why Does AI Feel More Expensive?](/articles/tokens-are-getting-cheaper-so-why-does-ai-feel-more-expensive) [2026-05-08](/articles/tokens-are-getting-cheaper-so-why-does-ai-feel-more-expensive) [// ai](/articles/tokens-are-cheap-thinking-isnt) ### [Tokens Are Cheap. Thinking Isn't.](/articles/tokens-are-cheap-thinking-isnt) [2026-04-24](/articles/tokens-are-cheap-thinking-isnt) ### Never miss an update Join 50,000+ developers getting our weekly tech insights. [YOUR EMAIL ADDRESS]() Subscribe --- # Why Technical Depth Matters More Than Content Volume > A reflection on why in-depth technical content is more valuable than superficial quantity in tech media. // Opinions 2026-03-21 [Jainil Prajapati](/authors/jainil-prajapati) 2026-03-21 - [The Noise Problem](#the-noise-problem) - [Deep vs. Shallow](#deep-vs-shallow) - [Shallow Content](#shallow-content) - [Deep Content](#deep-content) - [Why Depth Is Hard](#why-depth-is-hard) - [The Shravonix Philosophy](#the-shravonix-philosophy) - [Evidence of the Problem](#evidence-of-the-problem) - [What We’re Doing Differently](#what-were-doing-differently) - [Commitment to Testing](#commitment-to-testing) - [Primary Sources](#primary-sources) - [Correction Culture](#correction-culture) - [The Audience Question](#the-audience-question) - [Technical Depth as Service](#technical-depth-as-service) - [A Call to Creators](#a-call-to-creators) - [Conclusion](#conclusion) Content is being produced at unprecedented speeds in the tech industry. But speed doesn’t equal quality. ## The Noise Problem Scroll through any tech feed and you’ll see dozens of articles about the same topic: “ChatGPT in 2024,” “Why Python is Great,” “JavaScript Trends to Watch.” Most add nothing new. They’re derivative, superficial, often just AI-generated rewrites. This noise hurts readers. It wastes time and obscures genuinely valuable content. ## Deep vs. Shallow ### Shallow Content - Covers surface-level concepts - Repeats information already available elsewhere - Targets broad audiences with generic advice - Focuses on engagement metrics (clicks, shares) ### Deep Content - Explores technical implementations - Provides original analysis - Targets readers seeking understanding - Focuses on accuracy and value ## Why Depth Is Hard Writing deep technical content is difficult. It requires: 1. **Time to research**: Understanding a topic thoroughly takes hours, not minutes 1. **Technical expertise**: You can’t explain what you don’t understand 1. **Patience**: Slow, careful writing isn’t rewarded by today’s engagement algorithms 1. **Courage**: Admitting what you don’t know is rare ## The Shravonix Philosophy We’re not trying to be first. We’re trying to be correct, useful, and insightful. Every article we publish goes through this cycle: - Original idea, not topic hunting - Deep research into primary sources - Implementation testing (when applicable) - Multiple revision passes - Peer review from subject matter experts This slows us down. But the result is content that doesn’t expire. ## Evidence of the Problem I’ve personally encountered this countless times. Searching for “how to implement X” leads to: - Medium posts that clearly never tested the code - Tutorials that skip critical edge cases - “Beginner guides” that assume intermediate knowledge - AI-generated content that hallucinates features This wastes developer time. Time that could be spent building. ## What We’re Doing Differently ### Commitment to Testing When we write about code, we test it. When we write about tools, we use them. When we claim something works, we prove it. ### Primary Sources We link to documentation, source code, and original research. Other people’s coverage doesn’t count. ### Correction Culture When we’re wrong, we fix it. Publicly. With timestamps. This is rare in tech media. ## The Audience Question Some argue that deep content loses general audiences. That’s true. But that’s fine. Our audience isn’t “everyone.” Our audience is developers and technologists who want to understand things deeply. If that excludes some readers, it’s a trade we make deliberately. The internet has plenty of content for casual readers. It has very little for serious technologists. ## Technical Depth as Service Providing deep technical analysis is a service, not a content strategy. It helps people: - Make better technical decisions - Debug complex problems - Understand emerging technologies - Avoid common pitfalls This has real economic value. The time saved by accurate, detailed advice pays dividends. ## A Call to Creators To other writers and publishers: slow down. Your audience doesn’t need another hot take. They don’t need articles about why X is “the future.” They need things they can use. The internet has enough noise. What it needs is signal. ## Conclusion Technical depth matters. Not because it’s noble or virtuous, but because it’s useful. Shravonix exists to provide that depth. We’ll cover fewer topics than our competitors. We’ll publish less frequently than the content mills. But what we do publish will be worth reading. And that’s what matters. ## Comments ## Related Articles [// AI](/articles/mozilla-s-rebel-alliance-can-a-nonprofit-birth-a-startup-without-losing-its-soul) ### [Mozilla's Rebel Alliance: Can a Nonprofit Win the AI War?](/articles/mozilla-s-rebel-alliance-can-a-nonprofit-birth-a-startup-without-losing-its-soul) [2026-03-21](/articles/mozilla-s-rebel-alliance-can-a-nonprofit-birth-a-startup-without-losing-its-soul) [// ai](/articles/ai-in-india-2026-scared-leading-and-wasting-water) ### [AI in India 2026: Scared, Leading, and Wasting Water](/articles/ai-in-india-2026-scared-leading-and-wasting-water) [2026-05-16](/articles/ai-in-india-2026-scared-leading-and-wasting-water) [// ai](/articles/when-claude-became-my-relationship-coach-she-hated-ai) ### [When Claude Became My Relationship Coach (She Hated AI)](/articles/when-claude-became-my-relationship-coach-she-hated-ai) [2026-05-11](/articles/when-claude-became-my-relationship-coach-she-hated-ai) ### Never miss an update Join 50,000+ developers getting our weekly tech insights. [YOUR EMAIL ADDRESS]() Subscribe --- # Our Authors > Meet the writers and contributors behind Shravonix — a team of developers and technology experts delivering deep technical content. // Team ### Jainil Prajapati Founder & Editor Tech enthusiast covering web development, AI, and open source projects. ## // Want to Write for Us? We're always looking for new voices in tech. If you're passionate about technology and have something valuable to share, we'd love to hear from you. [Learn More](/guest) --- # Jainil Prajapati > Tech enthusiast covering web development, AI, and open source projects. // Author Founder & Editor Tech enthusiast covering web development, AI, and open source projects. [GitHub ](https://github.com/jaainil)[X/Twitter](https://twitter.com/shravonix) ## // Articles by Jainil Prajapati [// ai](/articles/ai-in-india-2026-scared-leading-and-wasting-water) ### [AI in India 2026: Scared, Leading, and Wasting Water](/articles/ai-in-india-2026-scared-leading-and-wasting-water) [India leads the world in enterprise AI at 80% but ranks 101st per person. The full honest picture: who's winning, what it costs, and what you should actually do.](/articles/ai-in-india-2026-scared-leading-and-wasting-water) [2026-05-16](/articles/ai-in-india-2026-scared-leading-and-wasting-water) [// ai](/articles/when-claude-became-my-relationship-coach-she-hated-ai) ### [When Claude Became My Relationship Coach (She Hated AI)](/articles/when-claude-became-my-relationship-coach-she-hated-ai) [I deleted 1,080+ messages on Claude's advice. Then she found out I was using AI to talk to her. Here's where I drew the line — and where I should've started.](/articles/when-claude-became-my-relationship-coach-she-hated-ai) [2026-05-11](/articles/when-claude-became-my-relationship-coach-she-hated-ai) [// ai](/articles/tokens-are-getting-cheaper-so-why-does-ai-feel-more-expensive) ### [Tokens Are Getting Cheaper. So Why Does AI Feel More Expensive?](/articles/tokens-are-getting-cheaper-so-why-does-ai-feel-more-expensive) [Prime's right that the AI economy is shifting — but the real story isn't about money. It's about GPUs, and nobody has enough of them. Let's break this down.](/articles/tokens-are-getting-cheaper-so-why-does-ai-feel-more-expensive) [2026-05-08](/articles/tokens-are-getting-cheaper-so-why-does-ai-feel-more-expensive) [// ai](/articles/tokens-are-cheap-thinking-isnt) ### [Tokens Are Cheap. Thinking Isn't.](/articles/tokens-are-cheap-thinking-isnt) [I put a daily limit on AI prompts. Not because I can't afford the tokens, but because my brain can't afford the attention. Here's why limiting daily AI prompts protects clarity.](/articles/tokens-are-cheap-thinking-isnt) [2026-04-24](/articles/tokens-are-cheap-thinking-isnt) [// ai](/articles/india-was-the-worlds-ai-warfare-lab-heres-what-actually-happened) ### [India Was the World's AI Warfare Lab. Here's What Actually Happened.](/articles/india-was-the-worlds-ai-warfare-lab-heres-what-actually-happened) [From the 2024 elections to Operation Sindoor, India faced the most complete AI information warfare campaign ever run against a democracy. Here's the full picture, technically. (177 chars — trim to: From deepfakes in 2024 elections to nuclear-risk AI disinfo during Sindoor — India is the world's most complete live case study in AI warfare. Here's the full technical breakdown. (181 chars — trim further:) China, the US, and AI in Indian elections — from deepfakes to Sindoor to arms market manipulation. The complete technical breakdown no one else is doing.](/articles/india-was-the-worlds-ai-warfare-lab-heres-what-actually-happened) [2026-04-21](/articles/india-was-the-worlds-ai-warfare-lab-heres-what-actually-happened) [// ai](/articles/qwen-just-became-the-most-downloaded-ai-model-heres-why-nobodys-talking-about-it) ### [Qwen Just Became the Most Downloaded AI Model — Here's Why Nobody's Talking About It](/articles/qwen-just-became-the-most-downloaded-ai-model-heres-why-nobodys-talking-about-it) [Alibaba's Qwen hit 1 billion downloads. 8 of 10 top AI models on Hugging Face are Qwen. Then the guy who built it walked out. Here's what actually happened.](/articles/qwen-just-became-the-most-downloaded-ai-model-heres-why-nobodys-talking-about-it) [2026-04-17](/articles/qwen-just-became-the-most-downloaded-ai-model-heres-why-nobodys-talking-about-it) [// ai](/articles/muse-spark-metas-new-ai-model-is-good-but-not-open-source) ### [Muse Spark: Meta's New AI Model Is Good. But Not Open Source.](/articles/muse-spark-metas-new-ai-model-is-good-but-not-open-source) [Meta Superintelligence Labs dropped Muse Spark after 9 months. Benchmarks are promising. But the open-source plot twist? Nobody's talking about it. Let's go.](/articles/muse-spark-metas-new-ai-model-is-good-but-not-open-source) [2026-04-09](/articles/muse-spark-metas-new-ai-model-is-good-but-not-open-source) [// ai](/articles/claude-code-source-leak-what-390k-lines-expose-about-ais-secret-sauce) ### [Claude Code Source Leak: What 390K Lines Expose About AI's "Secret Sauce"](/articles/claude-code-source-leak-what-390k-lines-expose-about-ais-secret-sauce) [Anthropic accidentally open-sourced their flagship AI coding tool via source maps. Here's what the leak reveals about agent architecture, internal features, and why developers are rewriting it in Rust.](/articles/claude-code-source-leak-what-390k-lines-expose-about-ais-secret-sauce) [2026-04-02](/articles/claude-code-source-leak-what-390k-lines-expose-about-ais-secret-sauce) [// ai](/articles/maven-smart-system-how-silicon-valley-optimized-the-kill-chain) ### [Maven Smart System: How Silicon Valley Optimized the Kill Chain](/articles/maven-smart-system-how-silicon-valley-optimized-the-kill-chain) [The Pentagon just made Palantir's Maven Smart System an official program of record. Here's how the AI stack behind modern warfare actually works — and why it matters.](/articles/maven-smart-system-how-silicon-valley-optimized-the-kill-chain) [2026-03-26](/articles/maven-smart-system-how-silicon-valley-optimized-the-kill-chain) [// tech](/articles/jio-hotstar-s-feature-flagging-how-they-ship-at-scale) ### [JioHotstar's Feature Flagging: How They Ship at Scale](/articles/jio-hotstar-s-feature-flagging-how-they-ship-at-scale) [JioHotstar ships \~12 features a week to 500M users without breaking 61M live viewers. Here's how their feature flagging system actually works.](/articles/jio-hotstar-s-feature-flagging-how-they-ship-at-scale) [2026-03-23](/articles/jio-hotstar-s-feature-flagging-how-they-ship-at-scale) [// Linux](/articles/open-claw-the-linux-of-ai-agents-or-a-security-nightmare) ### [OpenClaw: The Linux of AI Agents or a Security Nightmare?](/articles/open-claw-the-linux-of-ai-agents-or-a-security-nightmare) [OpenClaw hit 250K GitHub stars as the "Linux of AI agents." With CVE-2026-25253, 41% vulnerable skills, and 21K exposed instances, is it infrastructure or nightmare?](/articles/open-claw-the-linux-of-ai-agents-or-a-security-nightmare) [2026-03-22](/articles/open-claw-the-linux-of-ai-agents-or-a-security-nightmare) [// Opinions](/articles/why-technical-depth-matters) ### [Why Technical Depth Matters More Than Content Volume](/articles/why-technical-depth-matters) [A reflection on why in-depth technical content is more valuable than superficial quantity in tech media.](/articles/why-technical-depth-matters) [2026-03-21](/articles/why-technical-depth-matters) [// AI](/articles/mozilla-s-rebel-alliance-can-a-nonprofit-birth-a-startup-without-losing-its-soul) ### [Mozilla's Rebel Alliance: Can a Nonprofit Win the AI War?](/articles/mozilla-s-rebel-alliance-can-a-nonprofit-birth-a-startup-without-losing-its-soul) [Mozilla built a legal structure to stop itself from becoming the next OpenAI. Here's how the 80/20 split, PBC bylaws, and a $1.4B bet on open-source AI actually works.](/articles/mozilla-s-rebel-alliance-can-a-nonprofit-birth-a-startup-without-losing-its-soul) [2026-03-21](/articles/mozilla-s-rebel-alliance-can-a-nonprofit-birth-a-startup-without-losing-its-soul) [// IoT](/articles/iot-future-edge-computing) ### [The Future of IoT: Edge Computing and Beyond](/articles/iot-future-edge-computing) [Exploring how edge computing is revolutionizing the Internet of Things and enabling real-time processing at the network's edge.](/articles/iot-future-edge-computing) [2026-03-21](/articles/iot-future-edge-computing) [// Tech](/articles/apple-vision-light-the-ar-glasses-we-ve-been-waiting-for) ### [Apple Vision Light: The AR Glasses We've Been Waiting For?](/articles/apple-vision-light-the-ar-glasses-we-ve-been-waiting-for) [A comprehensive review of Apple's new Vision Light AR glasses and their potential impact on the augmented reality market.](/articles/apple-vision-light-the-ar-glasses-we-ve-been-waiting-for) [2026-03-20](/articles/apple-vision-light-the-ar-glasses-we-ve-been-waiting-for) [// AI](/articles/ai-models-2026) ### [Announcing Vite+ Alpha](/articles/ai-models-2026) [Announcing the alpha release of Vite+, the next generation of the Vite build tool with enhanced features and performance.](/articles/ai-models-2026) [2026-03-13](/articles/ai-models-2026) [// Tech](/articles/web-frameworks-benchmark) ### [VoidZero and npmx: Building Better Tools Together](/articles/web-frameworks-benchmark) [Exploring the collaboration between VoidZero and npmx to create better JavaScript development tools.](/articles/web-frameworks-benchmark) [2026-03-03](/articles/web-frameworks-benchmark) [// Tech](/articles/quantum-computing-breakthrough) ### [What's New in ViteLand: February 2026 Recap](/articles/quantum-computing-breakthrough) [A comprehensive recap of all the exciting updates and developments in the Vite ecosystem for February 2026.](/articles/quantum-computing-breakthrough) [2026-03-02](/articles/quantum-computing-breakthrough) [// Tech](/articles/tech-layoffs-stabilize) ### [Tech Industry Hiring Stabilizes After Tumultuous Year](/articles/tech-layoffs-stabilize) [Analysis of the tech hiring landscape in early 2026 and what it means for developers and companies.](/articles/tech-layoffs-stabilize) [2026-02-28](/articles/tech-layoffs-stabilize) [// Linux](/articles/rust-in-linux-kernel) ### [Rust in the Linux Kernel: One Year Later](/articles/rust-in-linux-kernel) [Reflecting on one year of Rust programming language integration into the Linux kernel development.](/articles/rust-in-linux-kernel) [2026-01-22](/articles/rust-in-linux-kernel) --- # Write for Shravonix > Contribute a guest post to Shravonix. Share your technical expertise with thousands of developers by writing about AI, Linux, IoT, or software engineering. // Contribute ## // Why Write For Us Shravonix welcomes contributions from the tech community. Whether you're a seasoned developer, researcher, or tech enthusiast, we want to hear your story. We focus on **technical depth** — articles that go beyond surface-level reporting to explore how technology actually works. ## // What We're Looking For 01 ### Technical Deep Dives Articles that explain how technology works under the hood — kernel internals, system architecture, algorithms, etc. 02 ### Industry Analysis Thoughtful commentary on tech trends, company strategies, and market movements. 03 ### Tutorials & Guides Practical how-to guides that help readers build real things or solve specific problems. 04 ### Opinions & Commentary Thought-provoking perspectives on technology's impact on society and industry. ## // Categories Linux Tech AI IoT Opinions ## // How It Works 01 ### Submit Your Idea Open a GitHub issue with your article proposal. We'll review it and get back to you within 48 hours. 02 ### Write Your Article Once approved, write your article in MDX format. We provide templates and guidelines to help you get started. 03 ### Review & Publish Our editorial team reviews your article for clarity and technical accuracy. Once finalized, we publish and promote it. ## // Ready to Contribute Have an idea? We'd love to hear it. Submit your article proposal via GitHub Issues and we'll get back to you soon. [Submit Article ](https://github.com/jaainil/shravonix-com/issues/new?template=GUEST_POST.md)[View All Authors](/authors) --- # Cookie Policy > Cookie policy for Shravonix - Learn about the cookies and tracking technologies we use. // Legal Last updated: January 1, 2026 ## Cookie Policy This Cookie Policy explains how Shravonix ("we", "us", or "our") uses cookies and similar tracking technologies on our services. Effective Date: 2026-01-01. --- ## What Are Cookies? Cookies are small text files placed on your device by websites you visit. They are widely used to make websites work more efficiently and to provide information to site owners. Cookies can be "session cookies" (deleted when you close your browser) or "persistent cookies" (remain on your device until they expire or you delete them). --- ## Types of Cookies We Use - **Essential Cookies** — Required for the basic functioning of our services. These cannot be disabled. - **Analytics Cookies** — Help us understand how visitors interact with our services so we can improve them. --- ## Other Tracking Technologies In addition to cookies, we may use the following tracking technologies: - web beacons - local storage --- ## Third-Party Cookies The following third parties may set cookies through our services: - **Umami** — Website analytics to understand visitor behavior ([Privacy Policy](https://umami.is/privacy-policy)) --- ## Your Consent - We display a cookie consent banner when you first visit our services. - You can manage your cookie preferences at any time via our preference panel. - You may withdraw your consent at any time; however, this will not affect the lawfulness of processing based on consent before its withdrawal. --- ## Managing Cookies Most web browsers allow you to control cookies through their settings. You can: - Delete cookies already stored on your device - Block cookies from being set on your device - Set your browser to notify you when a cookie is being set Please note that restricting cookies may impact the functionality of our services. Consult your browser's help documentation for instructions on managing cookies. --- ## Contact Us If you have questions about this Cookie Policy, please contact us: - \*\*Legal Name: \*\*Shravonix - \*\*Address: \*\*A/6 Shantivilla Residency, SVIT College Road, Vasad, Anand, Gujarat 388306, India - \*\*Email: \*\* --- # Privacy Policy > Privacy policy for Shravonix - Learn how we collect, use, and protect your data. // Legal Last updated: January 1, 2026 ## Introduction This Privacy Policy describes how Shravonix ("we", "us", or "our") collects, uses, and shares information about you when you use our services. Effective Date: 2026-01-01. If you have questions about this policy, please contact us at . --- ## Information We Collect We collect the following categories of information: - **Account Information** - Name - Email address - **Usage Data** - Pages visited - Browser type - IP address --- ## Data Retention We retain your data for the following periods: - **Account data**: Until account deletion --- ## Cookies and Tracking We use the following types of cookies and tracking technologies: - Essential cookies — required for the service to function - Analytics cookies — help us understand how the service is used --- ## Third-Party Services We share data with the following third-party services: - **Umami** — Website analytics to understand visitor behavior --- ## Your Rights You have the following rights regarding your personal data: - Right to access your personal data - Right to request deletion of your data - Right to correct inaccurate data - Right to receive your data in a portable format - Right to restrict how we process your data - Right to object to processing --- ## Contact Us Contact us: - \*\*Legal Name: \*\*Shravonix - \*\*Address: \*\*A/6 Shantivilla Residency, SVIT College Road, Vasad, Anand, Gujarat 388306, India - \*\*Email: \*\* --- # Terms of Service > Terms of service for Shravonix - The terms and conditions governing your use of our services. // Legal Last updated: January 1, 2026 ## Terms of Service These Terms of Service ("Terms") govern your use of services provided by Shravonix ("we", "us", or "our"). By using our services, you agree to these Terms. Effective Date: 2026-01-01. --- ## Acceptance of Terms You accept these Terms by: - using the service - creating an account --- ## Eligibility You must be at least 13 years old to use our services. By using the services, you represent that you meet this age requirement. --- ## Prohibited Uses You may not use our services for the following purposes: - Violating any applicable laws or regulations - Transmitting spam or malicious content --- ## Termination - We may terminate or suspend your access to our services at any time, with or without cause or notice. - You may terminate your use of our services at any time. --- ## Disclaimers - Our services are provided on an "as is" and "as available" basis. - We make no warranties, express or implied, regarding the reliability, accuracy, or fitness for a particular purpose of our services. --- ## Limitation of Liability - To the fullest extent permitted by law, we shall not be liable for any indirect, incidental, special, or consequential damages. - Total liability shall not exceed the greater of $100 or amounts paid in the past 12 months. --- ## Governing Law These Terms are governed by the laws of India, without regard to conflict of law principles. --- ## Changes to These Terms We may update these Terms from time to time. We will provide at least 30 days' notice via email or prominent notice on our website before changes take effect. Continued use of our services after changes constitutes acceptance of the revised Terms. --- ## Contact Us If you have questions about these Terms, please contact us: - \*\*Legal Name: \*\*Shravonix - \*\*Address: \*\*A/6 Shantivilla Residency, SVIT College Road, Vasad, Anand, Gujarat 388306, India - \*\*Email: \*\*