Anthropic is testing whether Claude Code can handle daily maintenance of the company's own software. In a few weeks, the AI created 388 pull requests, and 180 were merged after human review.
Claude has been running daily maintenance on Anthropic's in-house apps for "the last few weeks," according to Boris Cherny, the Anthropic engineer who created Claude Code. Instead of tying up developers with repetitive maintenance work, Claude handles it on its own. Cherny calls the results "surprisingly positive."
In his post, Cherny describes a setup where Claude, via Tag, runs daily routines through a dedicated Slack channel called "proj-claude-maintains-apps" across all of Anthropic's platforms: iOS, Android, desktop, web, CLI, and the Agent SDK.
Twelve routines cover the full range of code upkeep
Claude runs a battery of specialized maintenance routines, according to Cherny. A "Crash Fuzzer" opens apps in a simulator, taps around randomly to trigger crashes, analyzes the root cause, and creates a fix. A "Dup Unifier" scans the codebase for similar but slightly different abstractions and proposes merging them. The "Dead-Code Remover" strips out statically unreachable code and, for suspicious code, first adds logging so it can check the next day whether the code is actually unused.
| Routine | Function |
|---|---|
| Crash Fuzzer | Finds real app crashes and generates fixes for the root causes |
| Ant-only Shipper | Delivers or removes forgotten internal features |
| Logic Simplifier | Simplifies nested business logic |
| Logic Bug Fixer | Models complex logic to find and fix errors |
| Dup Unifier | Merges duplicate implementations |
| Dead Code Removal | Removes code that is proven to be unreachable |
| Useless Test Pruner | Removes tests that can never fail |
| Shipped-Feature Inliner | Removes feature flags for features that have been fully released |
| Flaky-Test Fixer | Analyzes and fixes unstable CI tests |
| Abstraction Improver | Simplifies overengineered abstractions |
| Abstraction Police | Fixes layer violations in the architecture |
Cherny shared some of his prompts in Slack, and there's no elaborate prompt engineering going on. He tells Claude in plain language to start daily routines for crash fuzzing on iOS, Android, and desktop, to use real apps without mocks, to trigger crashes, and to create pull requests with fixes.
388 pull requests created, just under half merged
Claude created 388 pull requests across Anthropic's repositories in the first few weeks, Cherny says. After a combination of automated Claude Code review and human review, 180 were merged, a rate of about 46 percent.
Claude usually gets the pull requests right on the first try, according to Cherny. When it doesn't, the team tweaks the routine so the AI does better the next day. That tuning sometimes takes a few days. Anthropic is now looking at ways to speed up the merge process for these kinds of mechanical changes.
More than half the auto-generated pull requests didn't make it through, which says something about both the limits and the potential of the approach. Cherny describes the experiment as "early signs of life" that autonomous AI-powered app maintenance could work.
Boris Cherny