The core problem
If you cannot evaluate the work, you fall back on proxies: the logo on a CV, confident communication, or familiarity with technologies you have heard of. All three correlate weakly with whether someone will be effective on your team.
Borrow a bar before you set one
Before interviewing anyone, get a practising engineer — an advisor, a contractor, an investor's CTO — to write down what "good" looks like for this role at this level. Concretely: which problems the person should work through unaided, and which are reasonably above the bar.
Without that, every candidate is judged against the previous candidate. That is how a search drifts for six months and ends with a hire nobody is confident about.
Ask for work, then talk about it
The most informative format we use: a small, realistic task at a normal pace, followed by a conversation about the decisions taken.
You are not grading the artefact. You are listening for whether the person can explain a trade-off, name what they would do differently, and disagree with you productively.
Signals that survive scrutiny
- Describes a failure they caused, and what changed afterwards.
- Asks about constraints before proposing a solution.
- Says "I don't know" without discomfort, then says how they would find out.
- Explains a technical decision to you, a non-specialist, without condescension.
Signals that do not
Speed at puzzles. Enthusiasm for your product. Willingness to accept the first offer. Each of these feels like evidence and predicts almost nothing.
Move faster than feels comfortable
Strong engineers hold offers for days, not weeks. A month-long process systematically selects from the group nobody else moved on quickly.
If you want to improve one thing about your hiring, compress the calendar before you add another interview stage.