Recognize the different forms of generated code
An inline completion, a chat-generated function, and an agent-authored pull request all produce code, but they expose different review surfaces. A function can be inspected locally; an agent change may also alter build files, dependencies, migrations, and deployment configuration across the repository.
Illustrative scenario: an assistant creates a CSV import endpoint. The endpoint is only one part of the feature. The reviewer must also inspect upload limits, error handling, permissions, parsing dependencies, and transaction behavior before deciding whether the implementation meets the product requirement.
Use generation where you can verify the result
Describe the contract before requesting code: accepted input, returned output, failure behavior, and constraints from the existing system. Provide relevant interfaces rather than the entire repository by default. Smaller tasks make it easier to compare the proposed change with an independent expectation.
Useful candidates include repetitive adapters, test fixtures, and a first draft of a well-specified function. Tasks involving unfamiliar authorization rules or production migrations need stronger domain review. Difficulty generating a result is different from difficulty proving that result safe and correct.
Keep ownership after generation
Read the diff as code your team will maintain. Verify that referenced packages exist, installed versions support the APIs, and the implementation follows local conventions. Document decisions that future maintainers cannot infer from the code, especially deliberate exceptions and externally imposed constraints.
Treat generated comments and tests as claims to inspect. A comment can describe behavior the code does not implement, and a test can reproduce an incorrect assumption. Require reproducible checks and a named reviewer before merging changes that affect users or stored data.
Working sheet
| Generated artifact | Main review question | Practical check |
|---|---|---|
| Helper function | Does it satisfy the contract? | Boundary input tests |
| Dependency change | Is the package necessary and supported? | Inspect package metadata and lockfile |
| Database migration | Can data remain consistent? | Rehearse against a sanitized snapshot |
| API endpoint | Who may invoke it? | Unauthorized and cross-user requests |
Questions about what is AI-generated code?
Is AI-generated code automatically low quality?
No. Judge the actual implementation against requirements and evidence. Its origin does not replace review.
Who is responsible for maintaining it?
The team that accepts and operates the change should assign ownership, just as it would for manually written code.
How should I use the working sheet?
Download the CSV and replace the sample rows with evidence from your repository. Record a clear acceptance check and an owner for each item. Keep unresolved assumptions visible rather than treating the example as a completed assessment.
Sources and further reading
Examples are illustrative. They are not customer results, security guarantees, or a substitute for checking the documentation for your installed version.
