Cursor and Claude now drive 66% of docs traffic. How are you keeping your docs from drifting

When your API docs fall out of date, it’s no longer just human developers getting frustrated. AI coding agents like Cursor and Claude Code are now consuming 50–66% of docs traffic. If your OpenAPI specs or Markdown guides drift from your actual codebase, these agents confidently hallucinate broken integrations—and developers simply abandon your API.

Code ships daily, but docs drift quietly. Usually, the docs task sits in a ticket nobody picks up until a customer complains or a support thread escalates.

I built DocFit (docfit.dev) to fix this by turning docs upkeep into a standard CI workflow. It’s a GitHub App that catches docs drift the moment code merges.

When a PR merges, DocFit:

• Understands the change: Reads the code diff, review comments, and the linked Jira/GitHub ticket to capture the actual intent, not just the code syntax.

• Locates the drift: Uses deterministic signals (OpenAPI diffs, code symbols) to pinpoint exactly which .mdx or .yaml files are now wrong.

• Proves the fix: Makes the minimal edit and runs your repository's native build and linters (like redocly lint or pnpm docs:build) in an isolated container.

• Leaves the decision to you: Opens a clean, verified docs PR. It never merges on its own.

It operates on a strict hierarchy: Deterministic signals first. AI judgement second. Verification last.

I am curious how other teams are handling documentation debt right now. Are you relying on docs sprints, blocking PRs until docs are updated, or just letting them drift?