The best AI lead is rarely the most technical person

The job is to decide what's worth doing, get people to change how they work, and keep the numbers real. Credentials help with the build. They don't get a sales team to use the thing.

By Matthew Steele3 min read28 October 2026

Two colleagues at a desk, one pointing at a printed process diagram while the other listens, a laptop open between them
Illustrative scene, not a customer's premises.

What the job actually is

Strip away the title and the AI lead does 5 things:

  1. Choosing: which jobs are worth automating and, more often, which aren't. Saying no to the shiny idea.
  2. Translating: turning "the quoting process" into something precise enough to build, and turning the tool's output back into something a salesperson will trust.
  3. Persuading: getting a team of busy and slightly suspicious people to change how they work, and keeping them changed while the first version's still clunky.
  4. Keeping the numbers real: insisting on a before number and an after number, and on someone checking the output.
  5. Protecting: knowing what data the tool may see and who may build, and stopping the quiet drift into risk.

Leadership and judgement cover 4 of the 5. One benefits from technical depth, and that one can be hired, borrowed or bought.

Why the most technical person often struggles in the role

This isn't a dig at technical people. It's what we've watched happen when the role goes to whoever knows most about the models.

They build what's interesting rather than what's expensive. The problem they understand best is the one in front of them, so the tools get built for their own department. They talk in tools, and the rest of the business tunes out. They measure accuracy and tokens when the business wants to know whether the quote went out a day faster. And because they can do it themselves, they do – and nothing gets handed over.

You end up with a very good tool that three people use.

Why the leader still needs a technical partner

The leader without credentials can't do it alone either, and the failure on that side looks different: tools chosen from a brochure, a prototype that never becomes reliable, and nobody who can tell when a vendor's overselling.

Who makes the best AI leadA chart with hands-on technical depth across the bottom and leadership and judgement up the side. Top right, rare: both. Top left, the usual best pick: strong leadership and judgement, enough hands-on use to be credible, paired with a builder. Bottom right, the trap: deep technical skill without the leadership to change how a team works. Bottom left: neither.Hands-on technical depthLeadership and judgementThe usual best pickHas changed how a team works before.Uses the tools every day, so is credible.Pairs with a builder for the technical depth.Can say no to a shiny idea.Rare: bothDeep technical skill and the standingto change how people work.If you have this person, you know it.Do not wait for one to appear.NeitherAvailable because they are not busy.The role becomes admin for the tools.The trapBuilds what interests them, not whatthe business needs. Speaks in tools.Measures the model, not the job.Nobody outside the project is using it.
Where the candidates sit. The usual best pick is strong on the vertical axis and credible on the horizontal one, with a builder alongside.

The pairing that works, in our experience, is a lead who uses the tools daily and a builder they can call. The builder might be one person inside the company who's good with systems, or an outside firm, or a mix. The lead decides, the builder makes it real, and neither tries to do the other's job.

One condition isn't negotiable. The lead has to be a daily user of the tools. A sponsor who's never written a prompt or corrected an answer will be led by the builder and won't notice.

The five questions we'd ask

  • Have you changed how a team works before, and did it stick?
  • Do you use these tools every day in your own job? Show us something from this week.
  • Can you explain, in plain words, something the tools get wrong?
  • Will people in other departments take a no from you?
  • Do you actually have the time, or will this get done in the gaps?

Three or more good answers and you've got your lead. The technical gaps are the easiest ones to fill.