Hire the question-asker, not the prompt engineer

Every second founder I talk to is trying to hire an "AI person" right now. Half the time they can't tell me what the person will do. They just know the board asked about AI twice and it feels like time somebody owned it.

So they write a job spec full of tools. Prompt engineering. RAG pipelines. Fine-tuning experience. And they end up hiring someone who's lekker with the machinery but has no opinion about whether the machine should be running at all.

Eish. The machinery is the easy part now. The models are good and getting better without your help, and the tooling changes every six months anyway. Whatever stack your new hire mastered last year is already two versions stale.

What doesn't go stale is judgement. The scarce person is the one who hears "we should add AI to onboarding" and asks what problem, exactly, we're solving. The one who does the arithmetic on what the demo costs at fifty thousand runs a month, out loud, in the meeting. The one who's comfortable saying "this one doesn't need a model, it needs a dropdown."

I wrote about the questions themselves in AI is a tool, not a strategy. This piece is about hiring the person who asks them without being told to.

The interview

Forget the take-home prompt test. Three questions tell you almost everything:

1. Ask them to kill an idea. Give them a plausible, exciting AI feature and ask what would make them cancel it. A question-asker names a failure mode and a cost within a minute. A tool operator starts describing the architecture.
2. Ask them to cost one. Take a feature that works beautifully in a demo and ask what it costs at a thousand times the volume. You're not marking the maths. You're checking whether the question feels natural to them or whether it's clearly never come up.
3. Ask what they shipped that had no AI in it. The best AI people have a track record of solving problems with boring technology when boring technology was the right answer. If everything they've touched since 2023 has a model in it, that's a pattern, and not a good one.

The uncomfortable bit

You might not need the hire at all. A fractional question-asker a day or two a week often beats a full-time tool operator, because the value is in the decisions, and there are only so many AI decisions a company needs to make in a week. The other four days you'd just be paying someone to fiddle with prompts.

But whether you go fractional or full-time, interview for the questions. The prompts, anyone can learn. The judgement about when not to write one, that's the whole job.

FAQ

What should I look for when hiring for AI?

Judgement over tooling. You want someone who can name the problem before naming the model, estimate cost at production volume without being asked, and point to things they shipped without AI when simpler technology was the right call.

Do I need a prompt engineer?

Probably not as a standalone hire. Prompting is a skill your existing engineers and designers can pick up in weeks. The rarer skill is deciding which problems deserve AI at all, and that's closer to product strategy than to engineering.

Is a fractional AI advisor better than a full-time hire?

For most small and mid-sized teams, yes, at least to start. The high-value AI work is a handful of decisions a month, not forty hours of weekly output. Buy the judgement part-time and let your existing team handle the build.