MokchiInterview Stress Testing

Posts

What a Bad AI Interviewer Looks Like

Mokchi

We are beginning to keep a list of things Mokchi should not do.

A bad AI interviewer:

Praises automatically

“Great answer” does not mean anything if every answer receives it.

Asks follow-ups without a reason

A follow-up should resolve uncertainty or test something meaningful.

Invents criticism

Difficulty is not the same as finding fault where none exists.

Forgets previous answers

A real interview is cumulative. Claims and contradictions matter across the conversation.

Confuses verbosity with depth

Longer answers are not inherently stronger.

Punishes missing information unfairly

If a candidate genuinely does not have access to a metric, the evaluator should assess how they reason about that limitation rather than inventing failure.

Becomes a coach mid-interview

The interviewer and answer coach are intentionally separate.

Mokchi will almost certainly do some of these things while we build it.

The important part is noticing when it does.

Follow the work

Occasional notes as we build Mokchi. No cadence promise, and we won’t use this for anything else.