The Age of Autonomous Code Has Arrived
Anthropic dropped a bombshell this week: as of May 2026, Claude now writes more than 80% of the code merged into its production codebase. Engineers are shipping 8× more code per quarter than the 2021–2025 baseline, and Claude Code’s creator Boris Cherny hasn’t written a single line of code by hand in eight months — he orchestrates tens of thousands of AI agents instead.
🔍 THE BOTTOM LINE: This isn’t incremental progress. It’s a phase change. When the company building one of the world’s most advanced AI models says its AI writes 80% of its own code and is calling for a global “pause button,” you should pay attention.
Anthropic reported that Claude’s success rate on complex, open-ended coding tasks hit 76% — up 50 percentage points in six months. The Mythos Preview model achieved a 52× speedup on AI training code optimization (where skilled humans manage 4×). Speaking at Fortune Brainstorm Tech, Cherny described managing agent fleets that range from “a few hundred to tens of thousands” on any given day, with subagent Claudes now doing the prompting of other Claudes — the human is increasingly out of the loop.
Inside the report: Anthropic’s own Anthropic Institute paper warns that recursive self-improvement “could come sooner than most institutions are prepared for” and explicitly calls for the world to develop a verifiable, coordinated pause mechanism.
For deeper analysis of Anthropic’s data, see our earlier piece: Anthropic’s Own Data Proves AI Is Already Building Itself.
NVIDIA Builds the Brain for Every Robot
NVIDIA went on a release spree at CVPR and COMPUTEX Taipei. The centrepiece: Cosmos 3, an open world foundation model that brings vision reasoning, multimodal generation, and action prediction together in one model. Think of it as a “world simulator” that lets robots, autonomous vehicles, and vision AI systems understand not just what they’re seeing, but what’s likely to happen next — then plan actions accordingly.
Why it matters: Physical AI has been bottlenecked by the cost and difficulty of real-world training data. Cosmos 3 generates synthetic video, robot-task data, and action trajectories (joint angles, gripper positions) from a single model, letting robots practice millions of scenarios without touching a real object.
Alongside it, NVIDIA released Nemotron 3 Ultra, a 550B-parameter Mixture-of-Experts model (55B active) optimised for orchestrating complex, long-running AI agents. This is the brain for the boss-level agent — the one that delegates to sub-agents across tools, environments, and time horizons.
“NVIDIA is building the compute fabric for the robotics era, and now they’re building the world model too.” — Singularity.Kiwi
AI-Driven Job Cuts Hit Records — 40% of All US Layoffs
The data is in, and it’s brutal. CNBC reports that AI has overtaken every other reason for US job cuts. In May 2026, 38,579 of 97,006 total layoffs were explicitly AI-linked — 40%. AI-linked layoffs for 2026 have already exceeded the combined total of 2024 and 2025.
Oracle is completing 30,000 job cuts by June 15 — the largest single layoff in the company’s history — explicitly tied to restructuring around AI and cloud compute.
Why it matters: This is a structural shift, not a cycle. The companies cutting jobs to fund AI investment aren’t planning to rehire those roles. Anthropic CEO Dario Amodei put it bluntly — “ideology won’t survive reality.”
For practical career advice in this climate, see our Career Compass: The Great Reshuffle.
Legal Hammer Falls: Florida Sues OpenAI
Florida became the first state to sue OpenAI, marking what legal analysts are calling AI’s “Big Tobacco moment.” The playbook comes directly from the social media lawsuits that produced $375M in settlements — courts have rejected Section 230 defences for chatbots, and over 20 lawsuits are now working their way through the system.
The University of Auckland’s Alexandra Andhov weighs in on whether OpenAI is putting profit over public safety, analysing the growing number of cases alleging harms from AI products.
Around the World in AI
xAI caught distilling Claude: Musk’s xAI trained its coding models on Claude outputs for months before Anthropic cut off access. A reminder that model distillation is the new intellectual property battleground.
South Korea plans AI Basic Society: President Lee Jae-myung announced a plan for universal AI access by year end, framing AI as basic infrastructure like electricity or internet.
Mathematicians warn of integrity crisis: The Leiden Declaration, signed by leading mathematicians, sounds the alarm on AI-generated research undermining scientific integrity.
❓ FAQ
Q: Is 80% AI-written code quality?
A: Anthropic claims Claude caught ~⅓ of bugs that caused past production incidents before they reached deployment. The 76% success rate on complex coding tasks is up from ~26% six months ago.
Q: What’s the difference between Cosmos 3 and Nemotron 3?
A: Cosmos 3 is a world foundation model for physical AI (understanding/reasoning about the real world). Nemotron 3 Ultra is an agent orchestration model (managing complex multi-step tasks across tools).
Q: Are AI-linked layoffs permanent?
A: The data suggests yes. AI has overtaken all other reasons for cuts, and companies aren’t signalling rehiring. The 2026 total has already exceeded 2024 and 2025 combined.
🔍 THE BOTTOM LINE: Code writes itself. Robots learn physics. Jobs shift permanently. The only constant in June 2026 is acceleration — and every institution is behind.