We let AI agents write 1,450 commits of our product. Here’s what code review became.

Description and use

Integrating AI agents into Deckgauge shifted code review from line‑by‑line to intent‑based checks, with strict CI gates ensuring only green commits reach humans, reducing review time and catching subtle logical errors.

Problem

Integrating AI coding agents into a measurement product risked introducing subtle, plausible bugs that traditional code review would miss.

Outcome

By enforcing gates before human review and focusing on data assumptions, Deckgauge reduced review time, increased confidence in metric correctness, and avoided subtle AI‑generated bugs.

Workflow

• 1. AI agent generates code and commits with co‑author trailer.
• 2. Automated pipeline runs lint, build, type‑check, and test suite; commit is blocked if any fail.
• 3. Human reviewer receives only green commits.
• 4. Reviewer asks three questions: (a) What data assumptions are made? (b) What happens at data boundaries? (c) Does the output match intended metric?
• 5. Reviewer approves or requests changes; changes are made by developer or AI as needed.

Evidence

We stopped reading diffs line by line and started reading them assumption by assumption.

Source

The discipline that actually made this work: no agent task reaches a human until it has passed lint, build, typecheck, and the test suite.

Source