The follow-up question is the interview

Watch a good technical interviewer and the pattern is always the same. They ask something broad. You answer. Then they pick one phrase out of your answer and ask about that. Then they do it again.

Three questions in, you are somewhere no preparation could have anticipated, discussing a specific thing you personally raised. That is not a hostile technique. It is the only reliable way to tell the difference between someone who has read about a thing and someone who has used it.

Why interviewers do this

Because first answers are cheap. Anyone can learn what a database index is, what a closure is, what eventual consistency means. Definitions are freely available and memorising them takes an evening.

What is not freely available is the second layer: why it works that way, what it costs, when it stops being true. That knowledge mostly comes from having been burned, and it is exactly what the follow-up is fishing for. An interviewer who only asked first questions would learn almost nothing about you.

How the drilling actually works

Follow-ups come in a small number of recognisable shapes. Once you can name them, they stop being a surprise.

• The definition probe. You used a term. Now define it precisely. Triggered by any word you said without unpacking.
• The mechanism probe. You said what happens. Now say why it happens, one level down.
• The boundary probe. When does this stop being true? What would break it?
• The cost probe. What did that choice lose you? Everything trades something.
• The alternative probe. Why not the other approach? Testing whether you chose or defaulted.
• The scale probe. What changes at a hundred times the volume?

You can rehearse against all six without knowing the topic. Take any answer you would give and ask yourself each of them in turn. Wherever you run out of road, that is a gap, and it is much cheaper to find it now.

Where the drill usually lands

There is a reliable pattern to which phrase gets picked. Interviewers drill on the words that carry the most meaning per syllable, because those are where a candidate is most likely to be borrowing language rather than describing understanding.

If you say a system is eventually consistent, expect to be asked what a user sees during the window. If you say you added a cache, expect invalidation. If you say you used a queue, expect what happens when consumers fall behind. If you say something is thread-safe, expect to be asked what makes it so. These are not obscure follow-ups. They are the obvious next question, which is precisely why they get asked.

Four things to do when the drill lands

1. Slow down rather than speed up

The instinct under pressure is to talk faster and cover more ground, hoping something lands. It reliably makes things worse, because more surface means more things to drill into. Take two seconds and answer the specific thing asked.

2. Answer the narrow question, not the broad one

A follow-up is deliberately narrow. Retreating to the general topic is a visible dodge, and interviewers read it as one. If you are asked what happens to in-flight requests during a deploy, talk about in-flight requests, not about your deployment philosophy.

3. Say where your certainty ends

"I know it holds the lock for the duration of the transaction. What I am less sure about is whether that is per row or escalates, and I would check that before relying on it." That sentence costs nothing and buys a lot, because it tells the interviewer your confidence is calibrated. Once they believe that, everything else you said gets more weight, not less.

4. Never invent a mechanism

This is the one genuinely unrecoverable move. If you make up how something works and the interviewer knows, the round is effectively over, and worse, everything you said earlier is now suspect. Reasoning aloud from what you do know sounds completely different from bluffing, and interviewers can tell the two apart easily.

What a good recovery sounds like

How to practise for an unpredictable question

You cannot rehearse the specific follow-up. You can absolutely build the depth it lands on, and you can practise the shape of a good response.

1. List every technology you would claim in an interview.
2. For each, write the answer you would give to the obvious first question.
3. Run the six probes above against that answer. Definition, mechanism, boundary, cost, alternative, scale.
4. Note where you run out. That is the real preparation list, and it is usually short.
5. Close those gaps by reading until you could answer two layers past them, then say it out loud once.

The out-loud part is not optional. Knowing something and being able to produce it under pressure, in order, in a sentence, are different skills, and only one of them gets tested. This is also why AI screens feel so much harder than they look: the follow-ups arrive without social brakes, so the drill goes one layer further than a polite human would push.