🧹Google Says Stale TODO Comments Are Derailing Its AI Coding Agents

Your old TODO comments are making AI agents dumber

Your AI coding agent is probably wasting time on bugs closed years ago. Google just published data showing that stale technical debt, like old TODO comments and obsolete bug references, pulls coding agents off track. They call it "agent distraction." If you're using Claude Code, Cursor, or any agent that reads your repo for context, this hits you directly. The agent doesn't know a TODO is dead. It reads it, treats it as a live instruction, and goes after it, even when your prompt never mentioned that bug. Outdated specs and lagging docs get ingested the same way, so the next person's agent inherits yesterday's confusion. The numbers are rough. About 13% of sessions exposed agents to inline TODO or FIXME references, and agents acted on them in 6% of those cases. Of those unprompted detours, 41% targeted tickets already closed and 57% targeted still-open bugs outside the request's scope. Each detour added 70% more steps, 77% more tokens, and 105% more wait time. Google's fix is scanner agents that comb the codebase, dig through years of git history to confirm both the TODO and its intent are dead, then open a PR to remove it. Over six months, the cleaner removed thousands of obsolete references with over 95% human reviewer acceptance.