How to hire your first AI engineer (without being one)

Most non-technical founders we talk to are about to make the same hiring mistake. They have heard "AI engineer" so they ask their recruiter for an AI engineer. The recruiter sends candidates with PhDs in machine learning. Those candidates are good at research. They are not the right shape for shipping production AI in your business.

Below is what to look for, what to ignore, and how to interview without pretending you can read code.

The shape you actually want

Your first AI engineer should be a senior software engineer who has shipped AI features in production. Not a researcher. Not a "prompt engineer" (this is not a real job for a senior role yet). Not a data scientist who has only built notebooks.

The skills that matter, in order:

1. Production engineering judgement — has shipped real software to real users for at least 5 years
2. Production AI experience — has shipped at least one AI feature that ran in production for 6+ months
3. Eval discipline — knows how to build and use evaluation sets, not just demos
4. Systems thinking — understands retrieval, caching, observability, cost management
5. Communication — can explain technical trade-offs to non-technical stakeholders

Notice what is not on this list: PhD in AI, papers published, deep theoretical knowledge of transformer architectures. None of those are wrong, but they are not predictive of who ships well.

What to ignore in the resume

"AI" as a buzzword

Every resume mentions AI now. Look for specific shipped AI features with names, scale, and outcomes — not "led AI initiatives."

Conference papers

Useful for research roles. Not predictive of production shipping. Many great AI engineers have never published a paper.

Specific model expertise

"Expert in GPT-4 fine-tuning" is a 12-month skill. Models change. What matters is the meta-skill: evaluating any new model against a benchmark and integrating it into a pipeline.

Five interview questions that work

1. "Walk me through an AI feature you shipped to production. What broke first?"

Real shippers always have a story about what broke. Researchers and prompt-engineers usually do not — because they did not ship to production.

Good answer: specific failure mode, what they did to fix it, what they would do differently. Hallucinations, retrieval failures, prompt injection, latency, cost overruns — all common stories.

Bad answer: "Nothing broke really, it worked great." Either lying or never shipped.

2. "How would you evaluate whether one AI model is better than another for our use case?"

Good answer: build a golden set of 50-100 representative examples, define one or two metrics that matter (accuracy, latency, cost), run both models, compare. Acknowledges that public benchmarks are usually misleading.

Bad answer: "GPT-4 is the best" or "I'd check the leaderboard." This is a researcher-style answer that misses production reality.

3. "Tell me about a time you decided NOT to use AI for something."

Good answer: a specific case where AI was the wrong fix — too expensive, too unreliable, the deterministic alternative was better. Demonstrates judgement.

Bad answer: "I always find a way to use AI." This is the person who will over-engineer your roadmap.

4. "How do you handle the case where the AI is wrong?"

Good answer: human-in-the-loop, escalation queue, confidence scoring, fallback paths, observability. The mature production-AI mental model. See our production AI properties.

Bad answer: "We tune the prompts until it stops being wrong." Misses the structural patterns.

5. "If you had three months to add AI to our business, what would you do first?"

Good answer: starts with discovery — what is the bottleneck, what data is available, what is the success metric. Picks one specific workflow. Acknowledges they need more context. See our rollout playbook.

Bad answer: launches into a long technical vision without asking about your business. This person will build the wrong thing fast.

Portfolio assessment without reading code

Ask for three things they have shipped:

1. A demo or screen recording showing the AI feature working
2. A write-up of what they built, why, and what they learned
3. References — someone non-technical from the team who can speak to delivery

Quality signals:

• The demo works on edge cases, not just the happy path
• The write-up acknowledges trade-offs and failures
• References describe the engineer as "easy to work with" and "explains things clearly"
• The engineer can tell you the production user count, not just "we shipped it"

Salary ranges (2026)

Mid-2026 market rates for the senior production-AI engineer shape:

• United States: $130-220k base + equity. Premium for ex-FAANG / ex-Anthropic / ex-OpenAI to $250-400k.
• United Kingdom: £80-140k base.
• India (Bangalore / Mumbai / Hyderabad): ₹35-90 lakhs base. Premium for ex-Big Tech to ₹1.2-2 cr.
• India (Tier 2 cities, remote): ₹25-60 lakhs base.
• Remote-first / global startup market: $90-180k base regardless of location.

Equity matters more at startups. Cash matters more at established companies. The right hire usually optimises for the work itself, not the package — so be honest about what the work will look like.

Where to source

• Your existing senior engineers' networks — the cheapest, highest-quality channel
• AI engineering communities — Latent Space, Hacker News, the local AI / GenAI meetups
• Open source — engineers who have shipped AI-related libraries or commits
• Specialist recruiters — only if internal sourcing has dried up; they are expensive
• LinkedIn search — filter on "shipped AI" / "production AI" in the description, not just AI in the title

Common mistakes when hiring

Hiring an AI researcher for a production job

Most researchers have never shipped to a real user. They build great prototypes. They struggle with deadlines, ambiguity, and the messy reality of production.

Hiring a junior with "AI bootcamp" on the resume

Bootcamps teach pattern-matching on toy problems. They do not teach engineering judgement. For your first AI hire, get someone senior.

Negotiating purely on salary

Senior AI engineers have options. The pitch that wins is interesting work + ownership + reasonable compensation, in that order.

Hiring before you have one clear AI workflow to ship

If you do not know what the engineer will work on for the first 90 days, do not hire yet. Run a project with a contractor first; figure out the workflow; then hire.

What this means for you

• Senior engineer with production AI experience, not researcher.
• Five interview questions surface judgement without code review.
• Portfolio + references + working demo before the offer.
• Salary is a market — pay it; do not low-ball your first AI hire.
• Read our operator playbook for the role this engineer enables.
• Read build vs buy AI first — you may not need to hire at all if you can buy.

Hiring your first AI engineer? Book a 30-minute call. We will help you sketch the role, the interview, and the comparison against just hiring us to ship the first feature for you.