Hiring your first AI PM

I get asked this by founders and by heads of product at companies that aren’t startups, and my first answer disappoints most of them.

You probably do not need one yet.

That’s not a general argument against the role. I think it’s a real discipline and I’ve written about what to screen for when you’re hiring into it. But the first AI PM hire is often made to signal seriousness about AI rather than to solve a problem anyone can name, and it goes badly in a specific and predictable way.

The market context, briefly

Useful to know what you’re competing with. Analysing 12,397 AI product roles posted in the US since January 2026, Axial Search found 47% at Manager level and roughly another quarter at Director level, with junior and mid-level individual contributor roles rare. 41% of postings came from companies with 10,000 or more employees. Median salary sits at $195,000.

Two caveats: it’s US data, and Axial is a recruiting firm publishing research about hiring.

The relevant point for a first hire is the seniority distribution. This isn’t a market where you will find an inexpensive person to figure AI out for you. It’s priced as an ownership role, and 41% of the demand is coming from very large organisations with deep pockets. If you’re a hundred-person company, you’re recruiting against that.

Which makes the question of whether you need one now a serious one.

The three signals that you do

You have more candidate use cases than you can build, and no way to rank them. This is the real one. A single AI feature doesn’t need a dedicated product manager. A queue of eleven ideas from six departments, with no shared basis for comparison, does. The job then is largely saying no, and that’s much easier for someone whose role is to.

A wrong answer has consequences somebody would care about. If your AI touches money, eligibility, health, employment or a regulated process, you need a person who owns the question of what happens when it’s confidently wrong. If the worst case is a slightly unhelpful suggestion, you can defer.

Somebody has to own evaluation continuously, and nobody does. Models drift, providers update, your data changes. If your answer to "who is watching quality in six months" is a shrug, that’s a role. I’ve written separately about how much of evaluation is product work rather than engineering work.

Two of three? Hire. One of three? Wait a quarter and look again.

The three signals that you do not

You have one feature and it’s working. Give it to whoever owns that product surface. Adding a specialist to a single working feature creates a person who needs to justify their existence, and the way they will do that’s by proposing more AI.

Nobody has sized anything. If no one can say what any of this is worth, an AI PM will spend their first two quarters doing the sizing work that should have preceded the hire, and you’ll conclude they’re slow. Do the sizing first, even badly. It may also tell you the hire is unnecessary.

What you actually want is an ML engineer. If the gap is building models, hire that. Product management is one of the most miscalibrated roles in hiring, with technical, growth and strategy profiles routinely confused, and adding AI compounds the problem. A product title on an engineering need produces a bad hire and an unfair performance conversation.

What the role owns, so you can write the description

Say the words in the job description, because vagueness here’s what produces the mismatch.

Which use cases get built and which are refused. What correct means for each one, and who arbitrates when experts disagree. What happens when the system is wrong, at the workflow level rather than the model level. The regulatory position — whether you’re the provider or the deployer of any given system, and what that obliges you to do. And ongoing quality after launch, by name.

Notice that none of that is model development, and all of it is decision-making. That’s the shape of the job.

On seniority, and the mistake I see most

The instinct is to hire mid-level for a first AI role — it feels like an experiment, and senior costs more.

I think that’s backwards, for a reason specific to this role. The core of the job is refusing things, and refusing things requires standing. A mid-level person who tells three department heads their AI ideas aren’t worth building won’t be heard, and will learn quickly to stop trying. You will then have someone who ships AI features nobody needed, which is the outcome you were hiring to avoid.

The market has largely arrived at the same conclusion, and that’s why so few of these roles are posted below Manager level.

The uncomfortable question to ask first

Before any of this: what happens if you do nothing for two quarters?

Sometimes the answer is that a competitor gets a real advantage, and then you should move. More often the answer is that you’d carry on shipping product and revisit it with better information, which is a perfectly good outcome and considerably cheaper than a hire made for the wrong reason.

The strongest position I can put this in: hiring an AI PM to work out whether you need AI is expensive discovery. Do the cheap version first.

Common questions

When should you hire your first AI product manager?

When at least two of three conditions hold: you have more candidate AI use cases than capacity and no shared way to rank them, a wrong output would have consequences somebody cares about, and nobody currently owns evaluation and quality after launch. With one of three, wait a quarter and reassess.

Do we need an AI product manager for a single AI feature?

Usually not. Give it to whoever owns that product surface. A specialist attached to one working feature acquires an incentive to propose more AI in order to justify the role, which is the opposite of what you want.

What seniority should a first AI PM hire be?

Senior. The core of the job is refusing use cases, and refusing requires standing that a mid-level hire won't have with department heads. Market data reflects this: across 12,397 US AI product postings since January 2026, 47% were Manager level and roughly another quarter Director, with junior and mid-level individual contributor roles rare.

How much does an AI product manager cost?

US market data puts the median at around $195,000, with 41% of postings coming from companies of 10,000 or more employees. That is US-only and from a recruiting firm's analysis, so treat it as an indication of the competition rather than a benchmark for other markets.

Should we hire an AI PM or an ML engineer?

If the gap is building or tuning models, hire the engineer. Product management is among the most miscalibrated roles in hiring, and putting a product title on an engineering need produces a poor hire and an unfair performance review. The AI PM role is decision-making: which use cases, what correct means, what happens when it's wrong, and who owns quality over time.

What should an AI product manager job description say?

Name the decisions rather than the technologies. Which use cases get built and refused, what correct means for each and who arbitrates disagreement, what happens when the system is wrong at the workflow level, the regulatory position as provider or deployer, and ongoing quality ownership after launch.

What should you do before hiring an AI PM?

Size at least one use case: what it's worth if it works, what happens if you do nothing, and what the current baseline is. That work has to happen anyway, it's cheap, and it sometimes shows the hire is unnecessary. Hiring someone to determine whether you need AI is expensive discovery.