Employer’s guide · Updated August 2026
How to hire AI engineers when you can’t judge the work
Vetting AI engineers is our core business — production evidence, references we call, specialties we verify. This is the interview loop we’d run in your seat, including the parts most companies skip and later pay for.
The short version: the AI talent market has a verification problem, not a supply problem. Anyone can claim AI expertise, demos take a weekend, and certificates screen for nothing — so the entire hiring process reduces to one discipline: verify production evidence at every step. Require shipped systems, interrogate the operating history, pay for a small working session, and call the references. Teams that do this hire well without deep in-house AI expertise; teams that interview on vocabulary hire whoever performs best in meetings.
The interview loop that finds builders
1. Portfolio screen (30 min, async)
Ask for one production system with their name on the operating history — what it does, who uses it, what broke. Demo links and certificate stacks are non-answers. This one filter removes most of the pool.
2. Systems conversation (60 min)
Walk through their system end to end: retrieval design, why these models, what the evals caught, cost and latency numbers. Builders answer with specifics and tradeoffs; performers answer with vocabulary.
3. Working session or take-home (2–4 hrs, paid if take-home)
A small, real problem from your domain — not a puzzle. What you are grading: how they define "working," whether an evaluation harness appears unprompted, and how they handle the ambiguity you left in the brief.
4. Engineering fundamentals (45 min)
AI engineering is engineering first. Ordinary code review, API design, debugging — at the level you would require of any senior engineer. Skipping this is how teams hire prompt-layer specialists into systems roles.
5. References, actually called (2 × 15 min)
Call the person who operated the system with them. Two questions do the work: "What did they own, specifically?" and "Would you put them on your next AI project?" The pause before the second answer is the data.
Total investment: about a week of calendar time per finalist. Compare that to the cost of the alternative — the industry is littered with six-figure engagements and $200k+ hires that a working session would have caught.
What good answers sound like
| Hire signal | Pass signal |
|---|---|
| Talks in evaluation criteria — "we knew it worked because…" | Talks in model names and buzzwords |
| Has cost and latency numbers for systems they ran | Has screenshots of demos they made |
| Says "it depends," then names what it depends on | Has one architecture for every problem |
| Asks about your data, users, and failure tolerance | Starts proposing solutions in minute five |
| Names what they don’t do | Claims the full stack, research to frontend |
The single most predictive signal is unprompted evaluation discipline— candidates who describe how they knew their system worked before users found out it didn’t. It is the least-taught skill in the field and the most screened-for by teams that have shipped. Role-scoped question sets live in our job description templates.
Set the comp and the route before you interview
Two decisions shape everything downstream. First, the band: US AI engineers run $150k–220k base at mid-level and $220k–300k senior, with infrastructure and GPU specialties well above — walk in anchored to the published salary bands or lose finalists at offer stage. Second, the route: for a first project, senior contract talent at $120–300/hr often beats a full-time search — you get proof on real work in weeks instead of a $300k+ commitment after months of recruiting. The sourcing options are compared in our hiring-routes guide.
And once the hire starts: give them a real problem in week one, a production deploy inside the first quarter, and the same evaluation-first standard you interviewed for. AI engineers with options leave companies that hire them into committee work.
Or start from a bench that’s already vetted
Every engineer in our network already cleared the bar this guide teaches — verified production deployments, references we called ourselves, a defined specialty. Tell us what you’re building and interview two or three matches instead of screening two hundred applicants.
Frequently asked questions
How do I hire an AI engineer if nobody on my team can evaluate them?+
Anchor the process in verifiable evidence instead of technical judgment you don't have: require a production system with an operating history, call the references who ran it with them, and pay for a small working session on a real problem before committing. If you still need technical depth in the loop, borrow it — a fractional AI leader or a vetted network that screens candidates for production evidence closes the gap.
What should an AI engineer take-home actually test?+
A scoped, real problem from your domain with deliberate ambiguity left in — graded on how the candidate defines success, whether they build any evaluation for their own output, and how they communicate tradeoffs. Two to four hours, paid. Avoid leetcode puzzles (wrong skill) and unpaid multi-day builds (senior candidates decline them, so you select for desperation).
What interview questions separate real AI engineers from pretenders?+
"Walk me through a system you shipped — what did the evals catch before users did?" "What did it cost to run, and what did you do about it?" "What would you not use an LLM for?" Real builders answer all three with specifics. The full screening-signal table and role-scoped questions are in our AI engineer job description templates.
Should I hire an AI engineer full-time or bring in contract talent?+
For a first project, senior contract talent is usually cheaper and faster: you pay $120-300/hr for exactly the weeks you need instead of carrying a $250k+ hire (a $300k+ real commitment) through three-plus months of recruiting. Hire full-time once there is standing AI work to own. The full math is in our salary guide.
How long does it take to hire an AI engineer in 2026?+
Three to six months through open-market recruiting for senior candidates — the pool with real production experience is small and heavily courted. Contract-first via a vetted bench compresses that to weeks, and converts to a hire once the fit is proven on real work.