--- title: "When an AI agent goes off the rails, file a bug to fix the documentation – then test the fix" url: https://daily.dev/posts/when-an-ai-agent-goes-off-the-rails-file-a-bug-to-fix-the-documentation-then-test-the-fix-rwzigopyu source_url: https://www.infoworld.com/article/4211198/when-an-ai-agent-goes-off-the-rails-file-a-bug-to-fix-the-documentation-then-test-the-fix.html type: article source: "InfoWorld" published: 2026-08-26T09:03:24.504Z updated: 2026-08-26T23:50:08.871Z tags: ["ai-agents", "mcp", "claude"] reading_time: 5 upvotes: 0 comments: 0 language: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # When an AI agent goes off the rails, file a bug to fix the documentation – then test the fix **[InfoWorld](https://daily.dev/sources/infoworld)** · 5 min read · 0 upvotes · 0 comments ## Summary An XMLUI co-maintainer describes treating documentation as a testable software component to support coding agents. When an AI agent fumbles a task because it can't find a relevant How To doc, that gap is filed as a bug and fixed by writing a new document, then verified via MCP search relevance scores (before/after comparisons showing improved match rankings). The approach scales through three signal sources: MCP query/result logs, mining application code across a portfolio of apps for common patterns developers search for, and session logs correlated with MCP logs to detect where agents struggled. The piece frames documentation writing as newly rigorous, measurable engineering work rather than purely editorial writing. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### How can I tell if my documentation is missing content that AI coding agents need? Log the queries agents make against a documentation MCP server along with the results and relevance scores they receive. Low relevance scores (around 0.7) on searches with no good match indicate a documentation gap; after adding a new document targeting that gap, the same query can score much higher (one case jumped to 4.2, six times higher than prior results), confirming the fix. _Developers refining agent-friendly docs can track practical workflow write-ups like this on daily.dev._ ### What's a good prompting technique to find missing patterns in a documentation catalog across multiple apps? Fan out subagents across a set of real applications and have them identify the top patterns a developer would search a How To catalog to find, then run those searches against the documentation MCP server to see which fail. This reverse-engineering approach treats existing application code as an inventory of patterns someone needed, surfacing documentation gaps at scale rather than one at a time. _Teams building agent-facing docs can follow workflow techniques like this via daily.dev._ ## Similar posts on daily.dev - [The LLM Flywheel Effect: AI That Writes and Tests Documentation](https://daily.dev/posts/the-llm-flywheel-effect-ai-that-writes-and-tests-documentation-975nvzbyu) · The New Stack · 2 upvotes · 0 comments - [You can just choose how many bugs you want now](https://daily.dev/posts/you-can-just-choose-how-many-bugs-you-want-now-yybwzht7p) · Read the Tea Leaves · 0 upvotes · 0 comments --- Tags: [#ai-agents](https://daily.dev/tags/ai-agents), [#mcp](https://daily.dev/tags/mcp), [#claude](https://daily.dev/tags/claude) [View this post on daily.dev](https://daily.dev/posts/when-an-ai-agent-goes-off-the-rails-file-a-bug-to-fix-the-documentation-then-test-the-fix-rwzigopyu)