# Why Stale CLAUDE.md and AGENTS.md Files Override Working Code in 90.1% of Runs (And the One-Sentence Fix)

Source: [DEV Community: python](https://dev.to/seanjhkwon/why-stale-claudemd-and-agentsmd-files-override-working-code-in-901-of-runs-and-the-10i2) · 2026-11-10 · Languages
Page: https://themorning.dev/story/why-stale-claude-md-and-agents-md-files-override-working-code-in-90-1-of-runs-an-7a9970527d2b6cb2
Kind: research · Impact: medium

## In brief

- Instruction files such as CLAUDE.md and AGENTS.md are usually written once when a repository sets up a coding agent and are rarely touched again.
- A study of 441 company repositories by Denisov-Blanch et al. (2026) found that 73.8% of AI configuration files were committed once and never updated.
- The source's headline claims stale instruction files override working code in 90.1% of runs; the excerpt does not detail how that figure was measured.

## Why it matters

If coding agents follow instruction files that no longer match the codebase, they may act on outdated guidance, so these files need to be maintained alongside the code.

## What to do

- Review CLAUDE.md and AGENTS.md when the codebase or workflow changes so agents do not follow outdated instructions.

## Coverage

- [Why Stale CLAUDE.md and AGENTS.md Files Override Working Code in 90.1% of Runs (And the One-Sentence Fix)](https://dev.to/seanjhkwon/why-stale-claudemd-and-agentsmd-files-override-working-code-in-901-of-runs-and-the-10i2) — DEV Community: python, 2026-11-10

Tags: ai-agents, coding-agents, claude-code, agents-md, developer-tooling, research

---

Brief generated by AI (claude-haiku-5-5) from the reporting listed under Coverage. Headlines belong to their publishers. The Morning Dev – https://themorning.dev