Interviewing engineers without the whiteboard theatre

Almost everyone who runs an engineering interview loop will tell you, privately, that the algorithm round is a weak predictor. Almost everyone runs it anyway, because it is legible, comparable across candidates and defensible when a hire does not work out.

What we replaced it with

A two-hour paid exercise on a real, small, self-contained problem from the actual codebase, with the candidate able to use whatever tools they normally use. Then a conversation about the choices they made.

The conversation is the signal. Anyone can produce working code for a small problem. The differentiator is whether they can explain the trade-off they took, name the thing they would do differently with more time, and recognise which parts of their solution are load-bearing.

Hire for the conversation about the code, not the code.

The objections, answered

It does not scale — correct, and it does not need to. If you are hiring six engineers a year rather than six hundred, two hours of senior time per finalist is cheap relative to the cost of a bad hire.

It disadvantages people with caregiving responsibilities — this is the serious objection. Pay for the time, keep it strictly bounded, and offer a synchronous alternative for anyone who prefers it.

• Pay for the exercise, always.
• Cap it at two hours and mean it.
• Use a real problem you have already solved, so you can compare approaches.