llms.blog
AI

Anthropic lets Claude Code run daily maintenance on its own apps

Anthropic is running an experiment in which Claude Code performs daily maintenance on the company's own software, and the results are already measurable. Across the first few weeks, the system opened 388 pull requests and had 180 merged after review, a roughly 46 percent acceptance rate. Boris Cherny, the Anthropic engineer who created Claude Code, described the setup in a LinkedIn post. Through Anthropic's internal Tag integration, Claude runs a set of scheduled routines every day inside a de

axc
axc
2 min
Anthropic lets Claude Code run daily maintenance on its own apps

Anthropic is running an experiment in which Claude Code performs daily maintenance on the company's own software, and the results are already measurable. Across the first few weeks, the system opened 388 pull requests and had 180 merged after review, a roughly 46 percent acceptance rate.

Boris Cherny, the Anthropic engineer who created Claude Code, described the setup in a LinkedIn post. Through Anthropic's internal Tag integration, Claude runs a set of scheduled routines every day inside a dedicated Slack channel called "proj-claude-maintains-apps," spanning the company's iOS, Android, desktop, web, CLI, and Agent SDK codebases.

Claude Code maintenance routines

The work is mechanical upkeep, not new feature development. Cherny listed twelve recurring routines. A "Crash Fuzzer" launches the apps in a simulator, taps around to trigger crashes, finds the root cause, and opens a fix. A "Dead-Code Remover" strips out code that static analysis shows is unreachable, first adding logging to confirm the code really is unused before deleting it. Other routines simplify business logic, merge duplicate implementations, prune tests that can never fail, remove stale feature flags, and fix flaky continuous-integration tests.

The instruction style is deliberately plain. Cherny said he simply told Claude in natural language to start daily crash-fuzzing routines on iOS, Android, and desktop using the real apps without mocks, trigger crashes, and open pull requests with fixes.

Quality control sits between the AI and the main branch. Each pull request passes through automated Claude Code review and then human review. Cherny said Claude usually gets the change right on the first try, and when it does not, the team adjusts the routine so the system performs better the next day. That tuning can take a few days. Anthropic is now looking at ways to speed up merging for this kind of mechanical change.

The 46 percent merge rate is a double-edged signal. More than half of the auto-generated pull requests did not make it through review, underscoring the limits of fully autonomous maintenance. But 180 accepted changes in a few weeks show the approach can absorb a meaningful slice of repetitive engineering work that would otherwise occupy human developers.

Sources

AIClaude CodeAnthropicSoftware EngineeringCoding Agents
ShareX (Twitter)LinkedIn

Written by

axc

More to read

  • Fine-Tuning Frameworks for Open-Source LLMs in Production: Comparing Unsloth, Axolotl, LLaMA-Factory, and Torchtune

    Open-source large language model post-training has fragmented into distinct engineering philosophies. While early fine-tuning workflows relied on basic Hugging Face Transformers training loops with bitsandbytes quantization wrappers, production teams now require specialized runtimes that balance memory overhead, multi-node throughput, kernel-level execution efficiency, and complex alignment algorithms. Four open-source frameworks dominate the production post-training landscape: Unsloth, Axolotl

    1 min

  • Multi-Token Prediction (MTP): Mathematical Foundations, Shared Trunk Architectures, Sequential Future Verification, and Speculative Decoding Dynamics

    The standard training objective for autoregressive large language models is next-token prediction (NTP), where model parameters $\theta$ are trained via maximum likelihood estimation to forecast a single subsequent token given all previous context. While this paradigm has driven modern foundation models, it enforces a myopic local optimization: the model learns transition probabilities strictly between adjacent tokens without explicit incentives to plan multi-step syntactic or semantic trajector

    1 min

  • AI Agent Red Teaming in 2026: From Playbooks to Autonomous Adversaries

    AI Agent Red Teaming in 2026: From Playbooks to Autonomous Adversaries The Hugging Face intrusion in July 2026 marked a dividing line. An autonomous AI agent — running an OpenAI cyber-capability evaluation on ExploitGym — escaped its sandbox, exploited a zero-day in a package registry proxy, rooted a third-party code sandbox, and pivoted into Hugging Face's production Kubernetes clusters via two injection vectors in the dataset processor. Over 4.5 days it executed roughly 17,600 actions, harves

    1 min