Claude Code Is Now 100% Written by Claude Code — 259 PRs, 497 Commits
@bcherny
When the team building the leading AI coding tool trusts it to write 100% of its own production code, the case for keeping AI at arm's length in your engineering org has effectively collapsed.
Claude Code Is Now 100% Written by Claude Code — 259 PRs, 497 Commits
Author: Boris Cherny (@bcherny) · Anthropic
The Headline
Claude Code has crossed a threshold that would have seemed like science fiction two years ago: every single line of code in the product is now written by Claude Code itself. Boris Cherny, the lead behind Claude Code at Anthropic, confirmed that 259 pull requests, 497 commits, 40,000 lines added, and 38,000 lines removed were all generated entirely by Claude Code — powered by the new Opus 4.5 model.
This is not a demo. This is a production codebase, shipping to users, maintained by AI.
New Built-In Skills
Alongside this milestone, two new built-in commands were announced — and Boris says he uses both daily.
/simplify
Runs parallel agents to improve code quality, tune efficiency, and ensure compliance with your CLAUDE.md configuration. Think of it as an automated code review and refactor pass that runs concurrently across your codebase.
/batch
Allows you to interactively plan complex code migrations, then executes them in parallel using dozens of agents — each running in an isolated git worktree. This means large-scale refactors or dependency upgrades that once took engineering teams days can be orchestrated and executed in a single session.
HTTP Hooks
A quieter but significant addition: hooks can now POST JSON to a URL and receive JSON back, rather than being limited to shell commands. This opens Claude Code up to richer integrations with external systems — logging pipelines, approval workflows, internal tooling — without requiring shell scripting.
Why This Moment Matters
The recursive nature of this milestone — an AI tool built entirely by itself — signals that agentic software development has moved from experiment to operational reality. The tooling (parallel agents, isolated worktrees, HTTP hooks) reflects the infrastructure patterns needed to make AI-generated code reliable at scale: isolation, parallelism, and auditability.
For engineering organizations watching from the sidelines, this is the clearest signal yet that the question is no longer whether AI can write production code, but how fast to build the workflows around it.