Founder's guide
Hiring your first AI engineer.
The title covers very different people. Get the type right, write the brief around the problem, and test for production experience, not vocabulary.
1. Decide which AI engineer you need
"AI engineer" means at least three different jobs. Hiring the wrong one is the most expensive mistake in this search, because each is strong where the others are weak.
- Applied AI or product engineer. Builds features on top of existing models: retrieval, agents, prompts, evaluation, and the product code around them. Most early-stage companies need this person first.
- Machine learning engineer. Trains, fine-tunes and serves models, and owns the data pipelines and infrastructure behind them. Right when your advantage is your own data or your own model.
- Research scientist. Works on problems with no known answer and publishes or prototypes new methods. Right when the research is the product. Rarely the first hire.
If you cannot say which of these you need, start with the problem: what should a user be able to do in six months that they cannot do today? The answer usually picks the type for you.
2. Write the brief around the problem
Strong candidates have choices. A list of frameworks will not move them; a hard, specific problem will. A good brief covers:
- The problem, in one paragraph, and why it matters to the business.
- What exists today: data, models, infrastructure, and who they would work with.
- What they would ship in the first three months.
- The must-haves, kept short. Three is plenty. Everything else is a nice-to-have.
- The salary band, the equity, and where the role is based.
Publishing the band saves everyone time. Candidates who are too far apart drop out early, and the ones who stay are already aligned.
3. Agree the scorecard before the first interview
Decide what you are scoring before you meet anyone, and score every candidate against the same list. For a first AI hire we would start with:
- Shipping. Has put AI in front of real users and kept it working.
- Evaluation. Measures quality with tests and metrics, not impressions.
- Data sense. Knows that data quality decides most outcomes, and has done the cleaning.
- Judgement. Picks the simplest approach that works, and can explain the trade-offs to a non-specialist.
- Ownership. Comfortable being the only person who understands the system, at least for a while.
4. Ask the five questions that matter
- Tell me about a model or AI feature you shipped that real users depend on. What broke after launch?Separates people who have run AI in production from people who have run it in a notebook.
- How did you know it was working?Listen for evaluation: test sets, metrics tied to the product, and how they caught regressions. "It looked good" is a no.
- When did you choose not to use a model?Good AI engineers reach for the simplest thing that works. If every answer is a bigger model, be careful.
- Walk me through the data behind your last project.Where it came from, how it was cleaned, what was wrong with it. Most of the job is here.
- What would you build in your first month here?Tests whether they understood your product and whether they can scope. You want a small, shippable answer.
5. Keep the process short
Good AI engineers are rarely on the market for long. Plan the whole process before you start and tell candidates what it is.
- An intro call, a technical conversation about their own work, a practical exercise, and a final meeting with the founders is enough.
- Keep any take-home small and close to your real problem. If it takes more than an evening, pay for it or replace it with a live session.
- Hold slots in the diary for each stage before the first interview, so nobody waits a week for the next step.
- Give feedback within a day or two of each stage. Silence is how strong candidates end up somewhere else.
Common mistakes
- Hiring a researcher when you need someone to ship product.
- Scoring on tools and buzzwords rather than shipped work.
- A brief that lists ten must-haves, so nobody fits.
- Leaving the salary band open, then losing the best candidate at offer stage.
- A process that changes halfway through the search.
Want help with the brief?
Send us what you have, even a few lines. We will turn it into a brief, check it with you, and tell you honestly what the market looks like for that role.
Ready to hire?
Let's talk about your next hire or your entire team build.