How do I audit why an AI-generated code change was made?

Keep a record that answers five questions. Who or what proposed the change? Which ticket or project asked for it? What evidence supported it? Which decisions were current at the time? Who made each decision that mattered?

Record the decisions behind the change.

Commit messages and pull request descriptions help, but they are not enough. People edit them, they leave out the decisions above the work, and they rarely say whether a decision was a draft or signed off.

Store the IDs and exact versions of the linked ticket and the decisions above it, Helen's flags, and what reviewers decided about each one. Tie the record to the exact version of the pull request that was checked.

  • Source Name the agent or person, the issue, and the ticket or project behind the change.
  • Sign-off Record which decisions someone signed off on as current and which were still drafts.
  • Evidence Keep the tests, the static checks, the written reasons from AI checks, and any checks that could not run.

Make the audit repeatable.

A later reviewer should be able to rebuild the decision from stored IDs and versions, without trusting a generated summary. Use summaries to find records, and treat the stored record as the source of truth.

Helen's Foundry links each pull request to its ticket and the decisions above it. Helen logs every flag she raises on those decisions and what a person decided about each one, so the reasoning behind a decision outlives the decision. Her flags are evidence, and a person makes the decision.

Ask whether you still want the change.

Who wrote the code matters less than two other questions. Does the company still want this change? Does the evidence justify shipping it?

See what Helen flags in your own plans.

Plans start at $20 a month.