Technical Interviews That Predict Actual Performance
Key takeaway
Assess candidates on work that resembles the job: reading and extending existing code, debugging something broken, and explaining a trade-off. Whiteboard algorithm puzzles predict interview practice, not delivery — and they disproportionately exclude experienced engineers who haven't revised for exams recently.
Hiring the wrong engineer is expensive and slow to correct; rejecting the right one is invisible and equally costly. Most interview processes are optimised for neither — they're inherited.
Assess what the job involves
- Reading code — give them a small, unfamiliar codebase and ask them to explain it and add a feature. This is what the job actually is on most days.
- Debugging — a failing test or a reproducible bug, worked through with them. It reveals method, which is the thing you're hiring.
- A design discussion on a realistic problem from your domain, including what they'd leave out and why.
- Communication: can they explain a technical decision to someone non-technical? For most roles this predicts effectiveness better than raw coding speed.
Practices that filter out good people
- Algorithm puzzles unrelated to the work, under time pressure, without a reference — a rehearsal exercise, not a skill test.
- Unpaid take-home projects longer than about three hours, which select for candidates with free evenings rather than ability.
- Panels where the candidate can't ask questions or look anything up. Nobody works that way.
- Trivia about syntax or standard library details that an editor would answer instantly.
- Five rounds spread over six weeks; the strongest candidates accept another offer before you finish.
Make it structured
- Same questions, same order, for every candidate for a role. Unstructured interviews mostly measure rapport.
- Define the scoring rubric in advance — what a strong, adequate, and weak answer contains.
- Interviewers write their assessment before discussing with each other, or the loudest opinion sets the score.
- Decide on evidence against the rubric, not on "culture fit", which is usually similarity in disguise.
Remember it works both ways
Good candidates are assessing you throughout. Tell them the format in advance, pay for substantial take-home work, give feedback quickly, and have them meet someone they'd actually work with. A process that respects their time is itself a hiring advantage — and it's the part of the process most companies never think to improve.