Most concerns about AI-written code come down to two questions: did anyone agree this was the right change, and did anyone check it actually works? Two checkpoints answer them.
Checkpoint one: approve the plan
Before any code is written, the work item is read, the codebase is investigated and an approach is proposed: what will change, where, and how it will be tested. An engineer confirms or redirects it.
This is the cheapest moment to change direction. Correcting a plan costs a comment; correcting a finished implementation costs a review cycle.
Checkpoint two: verify independently
After implementation, the repository's approved build, tests and checks run against the delivered code. The important property is who runs them.
| Question | Who answers it |
|---|---|
| Is this the right approach? | Your engineer |
| Did the build and tests pass? | The platform |
| Should this merge? | Your team |
The coding agent never grades its own work.
Why it matters for reviewers
A pull request that arrives with a confirmed plan and platform-run results is a different review task. Reviewers spend time on design and edge cases instead of wondering whether the change builds.
Getting started
Pick one repository and a handful of development-ready items, and run them through both checkpoints. Compare the outcome against your normal flow.