Assessment
What to look for in an AI engineer
The signals that separate engineers who ship production AI from those who have only experimented with it — and how to test for them in an interview.
Technical capability, ownership and communication — assessed consistently — predict success far better than a list of frameworks on a CV.
1. Production experience, not prototype experience
The clearest divider in the market is whether someone has run AI in front of real users, with real consequences.
- Has shipped a model or AI system that real users depend on, not only a notebook or proof of concept
- Can describe monitoring, evaluation and what they did when quality degraded
- Understands latency, cost per request and failure modes, not just accuracy
- Has owned something after launch rather than handing it over
Ask what broke in production and what they changed as a result. Engineers who have genuinely shipped answer this in specifics within seconds.
2. Ownership
Strong AI hires tend to take responsibility for outcomes rather than tasks.
- Chose the problem worth solving, not just the model
- Pushed back on a requirement when the data did not support it
- Built the unglamorous parts — pipelines, evaluation harnesses, data quality checks
- Can say what they would do differently with the same problem today
3. Communication across technical and business audiences
AI work fails commercially far more often than it fails technically, and communication is usually the difference.
- Explains a technical trade-off clearly to a non-technical stakeholder
- Frames work in terms of a business or operational outcome
- Sets expectations about uncertainty rather than over-promising
- Documents decisions so the team can maintain the system without them
4. Engineering fundamentals underneath the AI
The strongest AI engineers are strong engineers first. Frameworks change; the fundamentals do not.
- Comfortable with testing, version control, CI and code review
- Understands data modelling and where the data actually comes from
- Can reason about cloud cost, security and access control
- Picks the simplest approach that solves the problem
This is also the most transferable signal. Engineers strong on fundamentals move between model families and tooling generations without losing effectiveness.
How to test for it without a long process
You do not need five interview stages to assess this. You need the right questions in two.
- One structured technical conversation on a system they actually built, probing decisions and trade-offs
- One stakeholder conversation focused on communication and commercial framing
- A consistent scorecard across every candidate so comparison is possible
- Feedback within 48 hours, because strong AI engineers are typically in three to five processes at once
Key insight
Hire for what someone has put into production and how they think about trade-offs, not for the tools listed on their CV.
Planning a hire like this?
Tell us what you need to build and we'll help shape the hiring approach, timeline and market strategy.