All the latest on developer tools, open source, and building software.
The shift from autocomplete to autonomous agents isn't just faster coding. It's path dependency in action: once the industry crosses each threshold, the cost of going back exceeds moving forward. Here's the technical breakdown of where we are and the hard problems blocking what comes next.
Cloudflare rebuilt Next.js using AI in seven days (for $1,100). React and React Native moved to the Linux Foundation with proper governance. Hermes is coming to Node.js as a V8 alternative. Meanwhile, CSS Grid is landing in React Native and TanStack Router has an early native PoC. This week was unhinged.
Q1 2026 just closed and the AI industry's dirty secret is out: $505B in CapEx, 90% cheaper tokens, and creative accounting that would make Enron blush. If you're building on LLM APIs without serious value-add, your startup's about to get commodified into oblivion.
Goldman's 2026 AI memo isn't about innovation. It's a cost-cutting playbook dressed up as progress. When companies fire engineers to fund LLM subscriptions, they're not creating value. They're just moving payroll budgets to cloud providers while destroying institutional knowledge and compliance infrastructure.
Sockpuppetting hits 97% attack success on Qwen3-8B by prepending "Sure, here's how to..." to the model's output. No gradients, no optimization, just one line of inference code that outperforms GCG by 80 percentage points. The implications for self-hosted LLM deployments are wild.
97% adoption. Majority AI-generated code. Junior roles vanishing. This isn't incremental change, it's structural collapse of how we've built software for decades. Here's what actually changed and what it means for your job in 2026.
The Goldman Sachs memo wasn't a one-off. It's the playbook: overhire during boom times, then use AI as cover for mass layoffs. Executives are confusing cost-cutting with innovation, and developers are paying the price. Here's why this strategy is destined to collapse.
Story points map to complexity, not time. Except when we're cramming them into time-boxed sprints. They're relative estimates. Relative to what? Another inaccurate estimate? After 10+ years of rationalisations, it's time to admit the emperor has no clothes (or chickens).
Anthropic just accused DeepSeek of running 16M queries to steal Claude's capabilities. Except Claude was trained on millions of books, articles, and code scraped without permission. The irony is absolutely unhinged.
SQLite figured this out years ago: their code is public domain, but their test suite costs $120k. Now AI can clone your repo in minutes, but it can't replicate the years of edge cases and behavioural contracts encoded in comprehensive tests. The moat just shifted.
The AI industry says anyone can ship code now. What they don't say: you're legally responsible for every security hole the AI creates, even if you can't read the code it wrote. Courts don't care that Claude scaffolded your auth system. If it leaks PII, you're liable.
A solo dev built a viral AI agent framework by letting AI write the code. He doesn't review every line anymore. He just orchestrates. Welcome to the new job description: Code Agent Orchestrator. The glass is half-full if you're ready.