← All articles
AI Engineering3 min read

Your first AI hire isn't an engineer

The bottleneck in a first AI project is rarely model skill. Staff the judgement, the measurement, and the data first — then hire the engineer.

Your first AI hire isn't an engineer

Every company planning its first serious AI project writes the same job description first: senior ML engineer, five years of experience, production LLM systems preferred. It is the wrong hire to make first, and the job usually stays open for months while the project stalls.

The bottleneck in a first AI project is rarely model skill. It is everything around the model: someone who can say what "good" looks like for a given task, someone who owns the data the model will touch, and someone who will still be answering for the system six months after launch. An engineer cannot invent those things. They have to come from people who already understand the business.

The three roles that actually unblock delivery

The first hire that matters is a domain owner — someone from the business who can judge outputs. Not a data scientist, not a prompt engineer: the person who today decides whether a support reply, an invoice extraction, or a clinical summary is right. Without them, every eval is a guess, and the project drifts toward demos that impress in a meeting and fail on real cases.

The second is an evals owner. Somebody has to build the test set, run it on every change, and say no when a prompt edit or model swap makes things worse. This is not glamorous work, and it is the work that decides whether the system survives contact with production. In our experience, teams with a named evals owner ship slower for two weeks and then never look back; teams without one ship fast and spend months firefighting.

The third is a data steward — the person who knows where the data lives, who is allowed to see it, and what is wrong with it. AI projects die quietly on data access: three weeks waiting for a dump, a field that means something different in every system, a privacy rule nobody documented. A steward turns that from a project risk into a Tuesday task.

Then hire the engineer

None of this means engineers do not matter. They do — enormously. But an engineer joining a team that already has a domain owner, an evals owner, and a data steward can ship in weeks, because the questions that usually block them already have answers. An engineer joining a team with none of those spends their first quarter becoming all three, badly, while writing code in the gaps.

The pattern holds at every scale. The companies that ship AI fastest are not the ones with the most model expertise on day one. They are the ones that staffed the judgement, the measurement, and the data first — and then gave a good engineer a team that was ready for one.

We help teams staff AI projects for shipping, not for demos — starting with the roles that unblock delivery before a single model is chosen — because the fastest way to hire an AI engineer is to build a team they can actually succeed in.

Field note: A client once spent four months trying to hire a senior ML engineer while their AI project sat untouched. We suggested they pause the search and name a domain owner and an evals owner from inside the company instead. Within three weeks they had a working prototype with a real test set — built by their existing developers — and the engineer they eventually hired walked into a project with clear requirements, a way to measure success, and clean data access. The hire took six months. The shipping started in three weeks.

Related service AI products

Building something worth shipping?

We take on a small number of AI product engagements. Tell us what you are building — we reply quickly.