- At Anthropic’s Code with Claude conference in London on May 19-20, nearly half the attending developers said they had shipped a pull request in the past week that was completely written by Claude.
- Most of those who shipped such a PR said they did not read the code at all, per a show-of-hands led by Anthropic engineer Jeremy Hadfield.
- Hadfield said: ‘Most software at Anthropic is now written by Claude. Claude has written most of the code in Claude Code.’
- Code with Claude ran simultaneously with Google’s I/O 2026 keynote in Palo Alto, per the MIT Technology Review reporter.
What Happened
Nearly half the developer audience at Anthropic’s Code with Claude conference in London raised their hands when asked who had shipped a pull request in the past week that was completely written by Claude, MIT Technology Review reported. When asked who had shipped a Claude-written PR without reading the code at all, most hands stayed up. The two-day event kicked off on May 19 — the same day as Google’s I/O 2026 keynote.
Why It Matters
The conference audience was self-selected for Claude affinity, but the show-of-hands numbers crystallise how rapidly the developer workflow has changed since Claude 4 in 2025. Pull requests are the standard unit of software development — the fixes or updates to existing software that go through review before going live. They are the bread and butter that most professional developers spend their lives writing, or did until now.
Jeremy Hadfield, an Anthropic engineer on stage, said: “Most software at Anthropic is now written by Claude. Claude has written most of the code in Claude Code.” OpenAI, Google, and Microsoft make similar claims publicly. The category — coding agents that ship reviewable code at production quality — has matured from research project to default workflow in roughly 18 months.
Technical Details
The shift accelerated with Anthropic’s Claude 4.6 and 4.7 releases in February and April 2026. Claude Code, the company’s coding-agent product, is the surface where most of the developer-facing usage flows. Cursor’s Composer 2.5 release on May 17 — which matches Opus 4.7 and GPT-5.5 on SWE-Bench Multilingual at one-tenth the cost — provides an external benchmark validating the capability tier.
Code with Claude was the second year of Anthropic developer events, which also run in San Francisco and Tokyo. The London event ran May 19-20. Anthropic’s stated goal is to push automation as far as the underlying capability supports. Hadfield’s framing — “whether you like it or not” — captures the conference’s explicit tone about the speed of the shift.
Who’s Affected
Working software engineers face a fundamentally changed daily workflow. Code-review process, repository governance, security-scanning infrastructure, and the labour market for entry-level developer roles are all under structural pressure. Competing coding-agent products — GitHub Copilot, Codex, Cursor, Codeium, Replit, Continue.dev — face the same dynamic: capability has commoditised, the differentiation moves to UX, integration depth, and customer-specific tuning. Universities and bootcamps face the curriculum question raised by Stanford’s spring 2026 re-introduction of handwritten exams — how to credential developer competency when AI-assisted work is the default.
What’s Next
Anthropic’s developer events continue through Tokyo and San Francisco. Claude Code’s roadmap was sampled on stage but the specific product disclosures were beyond the scope of MIT Technology Review’s reporting. Industry watchers should expect parallel framings from Microsoft Build, Google I/O follow-on sessions, OpenAI DevDay (typically November), and continued competitive moves from Cursor, Codeium, and the open-source coding-agent ecosystem.
