The problem with trivia
Ask an engineer to define a database index and you learn whether they have read the documentation. Ask them about a query they had to make faster and you learn how they diagnose, what they measured, and what they traded away.
The second question is harder to prepare for and much harder to fake, which is precisely why it is worth asking.
Questions grounded in the candidate's own work
The strongest screening questions come from the resume in front of you. If someone lists a payments system, ask what happened when a payment partially failed. If they list Kubernetes, ask about the last time a deployment went wrong and how they found out.
- Walk me through a system you owned. What would you change if you rebuilt it?
- Tell me about something in production that broke. How did you find out, and what did you do first?
- What is a technical decision you argued for that turned out to be wrong?
- Where is the code you are least proud of, and what stopped you fixing it?
What a strong answer sounds like
Strong answers are specific about constraints. They mention numbers without being asked, name the tradeoff that was accepted, and describe what the alternative would have cost. They often include something that did not work.
Weak answers stay at the level of technology names. A candidate who says they used a queue to decouple services has told you a pattern. A candidate who says they used a queue because the downstream provider rate-limited them at 50 requests per second has told you about a real system.
Depth questions worth keeping
A small number of conceptual questions still earn their place, provided they probe reasoning rather than recall.
- How would you decide whether a slow endpoint is a database problem or an application problem?
- When would you deliberately store the same data twice?
- What breaks first when traffic increases tenfold, and how would you know?
- How do you decide something is worth testing?
Calibrating across candidates
The value of a question set comes from asking it consistently. Different questions for different candidates produce impressions rather than comparisons, and impressions are where bias does its quietest work.
Write down what a weak, adequate, and strong answer looks like before the first conversation. It takes twenty minutes and makes the difference between a ranked shortlist and a list of people you happened to like.