A correct answer in a technical interview tells you less than you think. Candidates can memorise patterns, and AI assistants can write working code in seconds. What's hard to fake is the reasoning behind a choice.
That's why follow-up questions matter more than the first question, whether a person or an AI interviewer asks them.
Why follow-ups beat harder first questions
Making the first question harder mostly filters for people who've seen that exact problem before. A good follow-up tests whether the candidate understands what they just wrote.
Compare:
- Harder question: "Implement an LRU cache with O(1) operations."
- Better follow-up: "You used a hash map plus a linked list. What breaks if this runs across 4 threads?"
The second one can't be answered from memory.
5 follow-up patterns that work
1. The trade-off probe
"Why this approach over the obvious alternative?"
Look for: the candidate names the alternative and a real cost (memory, readability, latency).
2. The scale shift
"What changes if the input is 100x larger?"
Look for: they identify the bottleneck before proposing a fix.
3. The failure case
"What input would break this?"
Look for: edge cases they find themselves, not ones you hint at.
4. The production question
"What would you change before shipping this?"
Look for: tests, logging, error handling, naming. These show real-world habits.
5. The rewind
"If you started over, what would you do differently?"
Look for: honest self-review. Defensiveness is a signal too.
How to score answers fairly
Write a simple rubric before interviewing anyone:
| Score | Meaning |
|---|---|
| 1 | Can't explain their own solution |
| 2 | Explains what, not why |
| 3 | Explains why and names one trade-off |
| 4 | Compares options and adapts when the constraint changes |
Use the same rubric for every candidate. Write down the evidence, not just the score.
Where AI interviewers fit
Running this for every applicant takes a lot of senior engineering time. Conversational AI interviewers can now ask these follow-ups automatically, adapting to each answer, and give the team a transcript to review.
The rules stay the same:
- Tell candidates they're speaking with an AI
- Give reviewers the full transcript, not just a score
- Keep a human on every advance or reject decision
Try it on one role
At CloudInterview we built conversational AI interviews around this approach: role-based questions, adaptive follow-ups and full transcripts for your team to review.
Top comments (0)