Engineering Leadership · Chapter 2 of 9
Hiring, and holding the bar
The highest-leverage thing a leader does, and the one most often done under pressure.
Every leader says they hire carefully. The test is what happens in month five of an open requisition with a team under pressure, because that is when bars actually get lowered and nobody records that it happened.
What a defensible process looks like
- A written bar, agreed before the first interview. What must be true, not a wish list. Without it, ‘good enough’ drifts with how long the seat has been empty.
- Structured interviews — the same questions, the same rubric, so you are comparing candidates rather than impressions.
- Independent write-ups before the debrief. If interviewers discuss first, the most senior or most confident voice anchors everyone.
- Someone empowered to say no whose objectives do not depend on the seat being filled — a bar-raiser, however you name it.
- A recorded reason for every hire and every rejection, because your future self needs to know whether your judgement is calibrated.
Hiring at scale, which is a different skill
Growing a team from ten to forty in a year is a specific competence and it gets probed hard in GCC and scale-up interviews. The failure mode is not finding people — it is that the culture and the bar are set by whoever joined in the last six months, because they outnumber the people who were there before. The controls: hire ahead of the onboarding capacity rather than behind it, keep tenured engineers on panels even when it is expensive, and slow down deliberately when the ratio of new to established gets uncomfortable.
The story you will be asked for
A hire who did not work out. Not the process — what you missed, and what you changed in your own interviewing as a result.
Everyone has one. A leader who claims they do not has either not hired much or is not being candid, and both readings are worse than the honest answer.
What actually predicts a good hire
The interview literature is clearer than the practice. Structured interviews — the same questions, in the same order, scored against a defined rubric — predict performance substantially better than unstructured conversations, and the gap is not close.
The reason is not that structure is magic. It is that unstructured interviews measure how much the interviewer enjoyed the conversation, which correlates with similarity to the interviewer rather than with future performance. Structure is a device for stopping yourself doing that.
- Define the rubric before you see a candidate, not after. A bar written while looking at someone is a bar written to fit them.
- Ask every candidate the same core questions. Without that you have no basis for comparison and every debrief becomes a contest of anecdotes.
- Score independently before discussing. The first person to speak in a debrief moves everyone else, and it does so more strongly the more senior they are.
- Separate signal from preference explicitly. ‘I would not enjoy working with them’ is sometimes real information and is more often a similarity judgement wearing a costume. Make people say which it is.
- A cross-team or bar-raising interviewer whose job is calibration against the company rather than the team is a genuinely effective mechanism, because a hiring manager under headcount pressure is not a neutral judge of their own bar.
The number everyone quotes about interview validity is out of date
The famous meta-analysis of selection methods — the source of every ‘structured interviews have a validity of 0.51’ slide — was substantially revised and corrected in 2022. The corrections lowered several headline validities and reordered the methods.
Most material still quotes the original numbers, and often attributes them with more precision than the underlying data supports.
The conclusions that survive are the robust ones, and they are enough: structure beats no structure; work samples and job-relevant tasks do well; unstructured interviews do poorly; and years of experience predicts very little once you are past the first few.
The practical version for a hiring manager: do not defend your process with a correlation coefficient. Defend it with the rubric, the calibration mechanism, and evidence that you check your own hires against the predictions you made about them — which almost no organisation does, and which is the only feedback loop that could actually improve a hiring process.
On take-home exercises: they are the most job-relevant format and the most socially expensive one, because they tax the candidates with the least free time — which is disproportionately people with caring responsibilities. If you use one, keep it under two hours, say so honestly, mean it, and pay for anything longer.
Check yourself