How to hire an AI developer
The field is two years old commercially, so credentials are thin and confident claims are cheap. These are the questions that separate people who have shipped from people who have read.
The short answer
Ask what they do when the AI is wrong. Anyone who has run a system in production will answer immediately and in detail — confidence thresholds, fallbacks, human review, monitoring. Anyone who has not will talk about model quality.
Ask how they measure whether a change improved things. If the answer does not involve a test set and a number, they are shipping on intuition, and you will be the one who discovers the regression.
- Ask what happens when the AI is wrong
- Ask how changes are measured before release
- Ask what runs in production today and for how long
- Ask who owns the code and where it is deployed
- Be wary of accuracy promises made before seeing your data
- Start with a small paid piece of work, not a large contract
Questions that separate experience from enthusiasm
Ask what they would measure, and how. Someone who has shipped will describe a baseline, a metric and a threshold. Someone who has not will talk about accuracy in the abstract.
Ask about a project that went badly and what they changed afterwards. Everyone who has done this more than twice has one. An answer that cannot produce a failure is a career too short or a memory too convenient.
Ask what they would not build for you. A supplier who says yes to everything is selling hours. The most useful answer we ever gave a prospective client was that a two-hundred-dollar-a-month tool already solved their problem.
Warning signs in a proposal
Accuracy claimed as a percentage with no measurement method attached. Timelines with no dependency on your side — every real project needs decisions and access from you. A fixed price for a scope that has not been written down. And model names used as though choosing one were the hard part.
Be equally wary of the opposite: a proposal so hedged that nothing is committed to. You should finish reading knowing what will exist, when, and how you will judge it.
Structuring the first engagement
Small, paid, and two to three weeks, ending in something running that you can judge for yourself. Not a discovery phase producing a slide deck — a working screen against your real data, however narrow.
That structure protects both sides. You learn whether they can actually build before committing a quarter's budget, and they learn whether your data and decision-making can support the project. Both of those are cheaper to discover in week two than in month four.
Frequently asked questions
Should we hire a freelancer, an agency or an employee?
For a first project, a freelancer or small studio is usually right: faster to start, cheaper to stop. Hire in-house once AI is core to the product and you need it maintained continuously. Agencies suit large multi-team programmes.
What are the warning signs?
Accuracy percentages promised before seeing your data. No mention of evaluation. Hosting the solution on infrastructure you cannot access. Recommending fine-tuning for a knowledge problem. Reluctance to say a project is a bad idea.
How do we check technical skill without being technical?
Ask them to explain a trade-off in their proposal in plain language. Real expertise simplifies clearly; borrowed expertise retreats into jargon.
What should a first engagement look like?
Small, paid, scoped to two or three weeks, ending in something running that you can judge. Both sides learn whether it works before anyone commits to a long contract.
Related pages
Tell us what you are building.
Send a short description of the problem and we will reply within one business day with an honest view of scope, cost and whether we are the right person for it.
Or email directly: contact@hire-ai-dev.com