Every company needs an AI team

Every time a new source of leverage arrives, companies grow a function around it. Industrialization gave us operations. The computer gave us IT — an entire department that had no reason to exist a generation earlier. The pattern is consistent: when something becomes both powerful and hard enough to use well, it stops being everyone's side responsibility and becomes someone's whole job.

AI is the next one. Most leaders sense this now, and the instinct is right. The trouble is what they do with it — they appoint a Head of AI and consider the box checked. That's exactly where it fails.

The two ways it fails

A Head of AI fails in one of two predictable ways.

The first is the silo. The company stands up an "AI team" that owns AI, and every other function quietly concludes that AI is handled now — it's their job, not mine. The work of the company goes on as before, untouched, while the AI team runs its own projects off to the side. A function that owns AI is the surest way to keep it out of the company. It's the same incoherence we've written about before: a pile of disconnected efforts instead of one organization that actually changed.

The second is the figurehead. A Head of AI is hired with a title, a mandate, and no team. They evangelize. They run pilots, commission decks, present a roadmap. But turning intent into change requires building things, and one person with a slide deck can't build. You get a champion without hands: plenty of motion, almost nothing shipped, and a year later not much that's different.

The structure that works threads between the two: a real, dedicated team whose explicit job is to make AI a capability every other role wields.

Enablement, not ownership

The mandate of an AI team is not to do the company's AI. It's to multiply the leverage of everyone else's work. Its success isn't measured by what it produces — it's measured by what every other team produces because of it.

The right mental model is good IT: infrastructure and enablement that quietly make the whole company more capable — not the ticket queue everyone learns to route around. Done well, you barely notice the function; you notice that everyone around it got faster.

This also reframes the silo problem. A team measured by its own output hoards; a team measured by what everyone else can now do is pushed to embed itself everywhere instead. That measurement is genuinely hard to get clean — we'll come back to it — but the incentive at least points in the right direction rather than the wrong one.

It has to get its hands dirty

The reason a lone Head of AI doesn't move anything is that enablement is built, not announced. The work is concrete, and it spans a real range:

• Implementing the tools that already exist where they fit — and getting people to actually adopt them, which is most of the battle.
• Building custom tools where off-the-shelf doesn't match how the company actually works.
• Training models on the company's own data where that data is the advantage no vendor can sell you.
• Wiring all of it into the systems and workflows people already live in, so the leverage shows up in the daily work instead of in a separate app no one opens.

This is implementation as much as strategy. Strategy is critical — knowing what to aim the organization at is the whole game. But an aim no one acts on changes nothing. A Head of AI who only sets direction produces plans; a Head of AI with builders produces change. You need the aim and the hands, and most companies fund the first and forget the second. The team that moves the needle is the one shipping into people's workflows every week, not the one presenting a roadmap once a quarter.

Measured in outcomes, not activity

A productivity function is dangerous to run on activity metrics, because activity is trivial to manufacture. More pilots, more dashboards, more "AI initiatives" — none of that is impact, and a team can look extremely busy producing exactly none of it.

The gains have to show up where they count: as measurable movement toward what the company is actually trying to do. We've argued that a modern company behaves like a function being optimized toward a goal. The AI team's job is to make that climb faster — to get the whole organization to its objective sooner. If the improvement isn't visible there, it isn't working, no matter how much got shipped. Be suspicious of the team that's always busy while the numbers that matter stay flat.

Where it sits

Because its job reaches into every function, the AI team can't be buried inside one of them. Tuck it under a single department and its mandate shrinks to that department's problems.

It needs to be its own function: a Head of AI with a small team of builders, reporting high enough to cross departmental lines — at the level of the other functions it exists to amplify.

That's the same box on the org chart as the silo we started with, and the opposite thing. The difference isn't where it sits; it's what it's measured on. The silo is judged by what it ships for itself. This team is judged by what every other team can do because of it. Same position, inverted mandate.

Senior enough to carry a real mandate, technical enough to ship against it. Small is fine. Toothless is not.

You don't have to build it alone

Most organizations don't have this talent in-house yet, and the people who can do the work well are genuinely hard to hire right now. Standing up the function — the first hires, the first builds, the operating model that makes it compound — is the hardest part, and the part where a wrong start costs you a year.

This is what we do. We help companies stand up their AI function: seeding it with people who've already done the work, shipping the first wave of tools that prove the model, and handing off a team that keeps compounding on its own. It's the same posture behind everything we build, including Damon — research-grade rigor carried all the way into production, not stopped at a slide deck.

Every company is about to need this function the way it needs IT. The ones that win won't be the ones that named a Head of AI first. They'll be the ones that built a team that made everyone else 10x — and could prove it.