How do you check a change written by AI?
Review note: The answer refers to the original author's CV or target job offer — adapt it to your own experience.
Answer out loud or in your head first, then compare.
Answer
I read the code diff and compare it with the goal. I check that the API, error handling or security haven't changed. Then I run checks suited to the risk and try a case that could disprove the solution.
For a search fix, I deliberately return the older response later. For a queue, I try a repeated message. It's not enough that the code ran on one easy input. I also check that the agent didn't remove an important test or add an unnecessary dependency. In the end, I can explain the key part myself and say what I didn't verify. A second review by a model can help, but it doesn't replace this process.
Why it matters: One of RAS's most important expectations of working with AI.
Possible follow-ups: What test would you write when a bug appears only with two concurrent requests?
How did it go? Be honest — “Not yet” puts the question on your review list.
Start with
- B-P14 · How do unit, integration and end-to-end tests differ? — JavaScript, TypeScript and Node.js
- I-P04 · How does a coding agent differ from a chat that suggests code? — Harness, MCP and coding agents
Further reading
- RAS Tech: Fair play in recruitment — how to use AI (in Polish)(opens in a new tab)Policy; accessed 22.10.2026. Transparency, review, accountability, confidentiality and the Codility condition.
- Karl Hörnlund / Canva: AI Interview Success(opens in a new tab)20.10.2025. An interviewer's first-hand account of planning, verification and owning decisions.