How to hire an AI engineer.

Every resume says AI engineer now. Very few of those people have put a model in front of real users and kept it running. This is how to tell the difference before you spend a quarter finding out.

The short version

• Screen for what shipped and survived contact with users, not for what was built in a notebook.
• Most of an AI role is engineering: data pipelines, evaluation, latency, and cost.
• Requiring a PhD narrows your pool to the most expensive and most contested candidates.
• If your data is unusable, you are hiring a data engineer and calling it an AI role.

What an AI engineer actually does all day

The modeling work is the visible part and the smallest part. An applied AI engineer spends most of their time on the systems around the model: getting data into a usable shape, building an evaluation loop so you can tell whether a change helped, controlling latency and cost, and handling the failure modes that only appear once real people are using it.

This matters for hiring because it means an AI engineer is a software engineer first. Candidates who can discuss architectures fluently but have never owned a service in production tend to struggle in the first ninety days, and the interview rarely catches it.

The questions that separate shipped from studied

Ask what they put in production and what broke. Anyone can describe fine-tuning. Far fewer can walk you through an evaluation set they built, explain why they chose those examples, and tell you what they changed when the numbers disagreed with user behavior.

Then ask about cost and latency. Engineers who have run something real have opinions about both, usually strong ones, because they have had to defend a bill or explain a slow response. Candidates who have only built demos have never had that conversation.

Finally ask what they decided not to use AI for. Judgment about where a model is the wrong tool is the clearest signal of someone who has actually shipped.

Where AI hiring goes wrong

The credential filter. Requiring a doctorate or a specific company logo feels like a safe proxy. It puts you in a bidding war for the smallest, most contested slice of the market while excluding applied engineers who have delivered more.

The mislabeled role. Many teams hire an AI engineer when the real blocker is that their data is scattered and untrusted. The hire spends a year building pipelines, both sides feel misled, and the person leaves.

Stale compensation data. AI salaries moved faster than most bands. Searches priced against last year's numbers stall at the offer after months of work.

No definition of success. Without a concrete six month outcome, the role drifts into open-ended research, which is expensive and the first thing cut when budgets tighten.

Hiring for regulated environments

In banking, insurance, and legal, model governance is not a nice-to-have. Explainability, audit trails, and rules about what data can touch which system shape what an engineer is allowed to build. Someone who has only worked where the answer was ship it and see will find those constraints slower than expected, and standard interviews do not surface it.

If you operate under those requirements, screen for them directly. Ask what approvals their last model went through and who signed off.

Frequently asked questions

Do I need a PhD to hire good AI talent?

Usually not. For applied work, engineers who have shipped systems often outperform researchers, and requiring a doctorate narrows the pool to the most expensive candidates on the market. Research depth matters when the work is genuinely novel, which is less often than job postings suggest.

What is the difference between an AI engineer and a machine learning engineer?

The titles overlap heavily and usage varies by company. In practice, machine learning engineer often implies more model training and evaluation work, while AI engineer increasingly means building products on top of existing models. Define the work rather than trusting the label.

How long does it take to hire an AI engineer?

Longer than general software engineering, because supply is tight and strong candidates hold multiple offers. Moving quickly between interview stages matters more here than in almost any other search.

Can AI roles be filled on contract?

Yes. Evaluation work, data pipeline builds, and proof-of-concept projects often suit contract engagements better than permanent hires.