Living recordslearning:interviewing-with-ai-as-a-review-loop
reviewpublicschema v1

Interviewing with AI as a review loop

When AI is allowed, the scarce skill moves from raw generation toward decomposition, verification, and repair ownership.

interviewsAI-assisted engineeringcode review

Current explanation

A practical AI-assisted interview loop is: state the contract, split the work, ask for a bounded candidate, inspect the diff, run the actual path, and explain which evidence changed your mind. The model is a fast collaborator, not the source of authority. Production readiness still includes tests, failure behavior, performance, accessibility, and explicit nonclaims.

Lesson path

  1. 01

    Frame

    current

    Write the contract and tool policy before prompting.

  2. 02

    Delegate

    next

    Ask for one bounded implementation slice.

  3. 03

    Review

    next

    Find one correctness risk the generated patch missed.

Open questions

  • What evidence distinguishes orchestration from unexamined code generation?

Selected Q&A

What should I narrate?

The ownership map, the test that can falsify your theory, and the reason you accepted or rejected the candidate.

Next actions

  • Run a 30-minute bug-squash rehearsal.
  • Practice rejecting a plausible but overbroad patch.

Sources

Revision trail

  1. · learning:interviewing-with-ai-as-a-review-loop@r1

    Turned current interview research into a repeatable review loop.