What do you do after an agent performs a wrong action?
Answer out loud or in your head first, then compare.
Answer
First I limit further harm: I disable the affected feature or tool and establish which tasks were affected. I check the real state in the system, because the agent's description may be wrong. Then I agree the fix and inform the right people.
Rolling back the code version may not undo a submitted request. You need to check, and sometimes correct by hand. Once things are under control, I reconstruct the cause, add a test and fix the missing check. I don't stop at “I wrote a better prompt”. If you haven't handled such a failure, answer as a scenario rather than inventing a personal incident.
Why it matters: Concrete ownership of a product after deployment.
Possible follow-ups: How do you check whether a similar error happened before?
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: Our Recruitment Process — All You Need to Know(opens in a new tab)Process; accessed 22.10.2026. A general description of the technical stage, internet access and AI; no guarantee of your round's format.