[ MSP.DD ] For investors and founders · Updated
How do you measure how much of a codebase was written by AI?
Short answer
Start with the commit history. Several AI coding tools sign their work: Claude Code adds a co-author trailer, Aider prefixes its commits, and coding agents commit under bot identities. Counting those markers gives a floor, not a total, because most AI-assisted code carries no marker at all.
Why it matters
The share of AI-written code tells you how much of the codebase may have shipped with less human understanding, and where to look first.
How to check
- 01Scan commit messages for co-author trailers that name an AI tool, such as "Co-authored-by: Claude".
- 02Scan author and committer identities for AI agent and bot accounts.
- 03Look for message conventions, such as Aider's "aider:" prefix and Claude Code's "Generated with Claude Code" footer.
- 04Treat the result as a minimum and confirm it in interviews.
- 05Compare the AI-marked share with test coverage and review practice.
Red flags
- A sudden jump in AI-marked commits right before a fundraise.
- AI-marked commits concentrated in security-sensitive code.
Good signs
- The team discloses AI use openly in commit trailers.
- AI-marked changes get the same review rate as everything else.
The numbers
- The open-source CHAOSS AI detection tool treats known AI bot committer emails and AI co-author trailers as high-confidence markers, and message patterns such as Aider's prefix as medium confidence. [1]