AI roles are not one role
"We need AI engineers" can mean five different searches. The first job of AI recruiting is to work out which one you are running.
- Research scientists and research engineers who design and train models, run experiments and read and write papers.
- Machine learning engineers who own training pipelines, fine-tuning, evaluation and getting models into production.
- Applied AI or AI engineers who build products on top of models: retrieval, agents, prompt and evaluation systems, and the product code around them.
- ML infrastructure and inference engineers who work on GPU clusters, distributed training, model serving, latency and cost.
- Data engineers who build the pipelines and labeling workflows that training and evaluation depend on.
AI companies also hire forward-deployed and solutions engineers, AI product managers, and the usual GTM, finance and people roles. Each of these draws from a different pool of people, with different signals of quality and different reasons to move.
Why the AI talent market is thin
Hands-on experience training large models, or serving them at scale, exists at a small number of companies and research groups. Most of the people with it are employed, well paid and not looking. Job titles do not help: "ML engineer" at one company is a research role, and at another it means maintaining a recommendation model.
So job posts and keyword searches return the wrong people. Good AI sourcing starts with a map: which teams have built the kind of system you are building, who worked on it, and what they have published, open-sourced or presented. Papers, GitHub repositories, conference talks and contributions to open models are often better signals than a résumé.
That research is the work of Embedded Sourcing. A dedicated sourcer maps the market for your role, sends outbound messages written for that specific person, screens for fit and hands your recruiters qualified candidates. Your team owns assessment and decisions.
What AI candidates want to know
Senior AI candidates choose between frontier labs, large tech companies and well-funded startups. They ask harder questions than most candidates, and they ask them early.
- How much compute the team has, and who decides how it is used.
- Whether the company trains its own models, fine-tunes open models or builds on third-party APIs, and how that might change.
- What data the team can access, and how good it is.
- The split between research and product work, and whether they can publish.
- Who is already on the team, and how much runway the company has.
A recruiter who cannot speak to these topics loses the candidate in the first call. Your recruiter needs a real brief from your technical leaders and the judgment to know when to bring them into the conversation directly.
Offers are often equity-heavy and compared against large cash packages, so the close needs the same care as the search.
Interview loops for AI and ML roles
A generic software loop misses what matters in AI hiring. Strong loops for these roles usually combine:
- A coding round that reflects the actual work, such as data manipulation or model code, rather than only puzzle questions.
- ML fundamentals and an ML system design round: how the candidate would train, evaluate, serve and monitor a model for your problem.
- A deep dive on past work or a research presentation, where the candidate explains decisions and tradeoffs.
- For applied roles, a practical exercise on evaluation, retrieval or a product problem.
Calibrate research and applied candidates separately. A strong research scientist and a strong AI engineer will not look alike on the same scorecard.
Keep the loop short. Candidates in this market usually have several processes running, and a two-week gap between rounds often ends a search. Your recruiter tracks every step and keeps hiring managers on schedule.
How Mavericks supports AI companies
Most AI teams start with Embedded Sourcing, because finding the right people is the hard part. When you also need screening, interview management and offers handled, Embedded Recruiting covers the full search. Senior leaders, such as a head of research or VP of engineering, go through Executive Search.
Our recruiters work with Mav, the AI companion every Maverick uses day to day. People run the search, build the relationships and make the judgment calls.
Teams Mavericks has supported in AI and machine learning include Writer, AssemblyAI, Twelve Labs, FriendliAI, Exa, Harmonic and EliseAI. We also support data and infrastructure companies such as Airbyte and Onehouse, where many of the same engineers work.
For broader software hiring, see technology recruiting. If you need to add capacity across many roles at once, see recruiting capacity.