← Blog

Building an AI team: who to hire first

I’ve helped companies build data science teams many times, from one person to a small group. Most of them now call it an AI team, and the job titles have changed. The way to build one hasn’t changed much: start small, hire carefully, and grow around one strong first hire.

I wrote about this across several posts between 2017 and 2020. This brings them together.

First, make sure you need the hire

An AI hire is expensive, hard to find, and hard to manage if the work isn’t defined. Before you write the job description, try what I call visioning. Spend ten minutes writing, in as much detail as you can, what will be different once this person is on your team. What will they work on in their first few weeks? Which metric will that work change?

Some answers mean you’re ready:

  • “We recommend our most popular products to everyone. We’ve been collecting purchase data for years, and we’re sure we’re leaving revenue on the table. We just don’t know where.”
  • “We have months of customer and order data and no way to estimate lifetime value or predict who’s about to leave.”

Some answers mean it’s too early:

  • “We’re about to start collecting data, and we need someone to tell us which fields will matter.”
  • “Our chief executive read that AI is growing fast, and we don’t want to be left out.”
  • “We have a big pile of data. There has to be something in there.”

The last one is the most common, and it’s the hardest situation for any new hire to walk into.

Sometimes you don’t need an AI hire at all. If what you really need is someone to run reports and keep the company informed about what’s changing, hire an analyst and save a lot of money. If you’re not sure, I wrote a separate guide on whether your company needs AI.

Your first hire

If one person will carry your AI work for a while, hire an experienced generalist who:

  • Can explain the work, and its limits, to people who aren’t technical
  • Has worked in your industry, or a close one
  • Has worked with your technology team to get data and clean it
  • Has put simple models into production

I can’t stress communication enough. In the interview, ask the candidate to explain a project on their resume. You should understand it right away. If they struggle now, imagine them under deadline, explaining a last-minute change to your developers or a revenue drop to your chief executive.

The next two hires

Once the first hire is settled, I recommend adding two roles. I call the result the Magic Three:

  1. Senior AI engineer (your first hire). Turns business goals into clear objectives for the team, builds the models, and coordinates the work.
  2. Data engineer. Makes the data available, and builds and runs the pipelines that move and clean it for the whole team.
  3. Analyst or junior engineer. Supports the senior engineer day to day, helps shape the pipelines, and answers questions from the rest of the company.

Data engineers are worth more than most companies expect. Often the next hire after these three is another data engineer, because cleaning and moving data is where most of the time goes. I wrote about that in why so much of an AI project goes to cleaning data.

What experience is worth

I see stories like this all the time: “Two years ago I couldn’t write code. Now I’m an AI engineer.” Or a twelve-week course followed by applications for senior roles.

I’m glad the field is growing, and I’ve hired and worked with some very bright people who came from short courses. That’s no fault of theirs. Real data is messy, full of gaps and inconsistencies, and it takes years to learn how to handle it. Senior engineers need much less oversight because they’ve already made the common mistakes.

In interviews, I look for three things:

  • One hard-won project. Ideally the data was a mess before the candidate got involved. I want to hear what they did about it, why the project mattered, and what the hardest part was.
  • Real experience on several projects, outside of coursework.
  • Curiosity and self-reliance. So much of this job is figuring things out alone.

Skip the job posting that lists 50 technologies. Ask candidates to walk you through a project they care about. Tools change. Fundamentals don’t.

When to bring in outside engineers

Building a team takes time, and even a strong first hire needs time to learn your data before the work pays off. Many companies start with outside engineers who work inside their team, then hire as the work proves itself.

That way of working has caught on. Monthly job listings for forward-deployed engineers, who work inside a customer’s team to put AI into production, rose 800% between January and September 2025, according to First Round Review. We have worked this way since 2017. You can read how in how we work with clients.

If you go that route, the questions in how to choose an AI consulting firm will help. And plan the hand-off from the start. With Bennett ONE, your own engineers can take over named parts of the system on a schedule you set, as the Bennett ONE questions and answers explain.

Talk to us

If you’re deciding between hiring and bringing in help, the first call with one of our senior engineers is free and takes 30 minutes. We’ll tell you plainly which makes more sense.

Let's talk about what you want to improve

Talk to an engineer