How building software is changing at Anthropic
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Anthropic's Claude Platform team spent six months building Claude Managed Agents, a production agent harness launched in April that runs on Anthropic-managed or customer-owned infrastructure.
- Jarred Sumner, creator of Bun, migrated 500,000+ lines of code from JavaScript to Rust in under two weeks using AI tool Fable and $165,000 of tokens — a project the article notes previously would have taken a small team a year.
- The Claude Platform layer is being migrated from Python to Rust because Python's single-threaded design can't handle the API's performance demands at scale.
- At the 3,500-employee lab, code review and testing are increasingly AI-driven, prototyping is more fluid, and verification now takes longer than implementation.
- Anthropic teams now operate with a maximum of two engineers per project, while two-pizza teams, PRDs, and planning rituals persist as unchanged practices.
- Claude Managed Agents required mid-stream re-architecting — decoupling the "brain" (Claude and harness), "hands" (sandboxes and tools), and "session" (event log) into separate interfaces, plus a vault and credential abstraction.
- Anthropic engineers told the author that the more hands-on they are with AI tooling, the less they fear their jobs will be replaced.
Why it matters: The Bun rewrite — 500,000 lines ported in 11 days for $165,000 of tokens — demonstrates that AI-assisted large-scale migration is now economically viable where it wasn't before, compressing timelines that previously made such rewrites impractical. For engineering orgs watching AI labs, Anthropic's experience signals a split: foundational practices (two-pizza teams, PRDs, planning) survive, but code review, verification effort, and project staffing ratios are being actively reshaped.
