Extend a feature in a Python codebase you have never seen, using an AI assistant, and explain every change it made.

How to answer

A Blind poster wrote, in September 2026, asking for a friend with an upcoming Cohere FDE onsite, that one round is AI-enabled coding on CodeSignal of about 45 minutes, and that the recruiter said “they give you an existing python codebase and ai”. ^[1]Source 1Cohere FDE onsite, AI enabled coding round?PublisherBlindSource typecandidate report on Blind The onsite had not happened yet, and the poster did not say how the round is judged. Own the diff: the assistant writes it, you answer for every line.

1. Read before you prompt. Run the tests, trace the path your feature will take, and name the pattern the change should copy: “Bad parameters raise FilterError, and the handler turns it into an HTTP 400.”
2. Write the acceptance tests yourself. Include the case you expect the assistant to miss. Your tests are the spec, so they must not share its assumptions.
3. Prompt narrowly. Name the files, the pattern to follow and what not to touch. Ask for a plan first if the change spans files.
4. Read the diff hunk by hunk, out loud. For each one, say what it does, why it is needed and what it assumes. Then keep it, fix it or revert it. Revert anything you did not ask for or cannot defend.
5. Check what it relied on. Library behavior in the repo’s Python version, helpers it calls, how the data is really stored. Run your tests, not only its own.
6. Close with a summary of the diff. Say what changed, what you rejected and why, and the question you would take to the customer.

The trap is accepting a diff because its own tests pass, then describing it in general terms. The interviewer will point at one line and ask why it is there.