Essay · August 2026

AI Is a Capability, Not a Department

AI is not a thing you can fence off and assign to an owner.

When a technology this big arrives, the instinct is to make it a thing: stand up an AI group, give it the mandate, and point everyone else at it. The instinct is understandable, and it rests on a category error. AI is a capability. It lives in the work, in the hands of the people doing it, which is exactly why you cannot fence it off and hand it to an owner.

Think about the capabilities you already depend on. No one owns the spreadsheet. No one stood up a department to govern email. They are woven into how everyone works, and the standards that keep them safe are built into the work rather than bolted on by a separate team that keeps them at arm's length. AI is heading to the same place, faster. The companies that treat it as something everyone uses will get there; the ones that treat it as a unit to stand up will spend years pointing at a team while the rest of the organization waits.

The moment AI becomes one group's job, two things happen. The group turns into a bottleneck, because every other team waits on it to build, approve, or bless. And everyone else stops treating AI as theirs, since it is now someone else's lane. You have concentrated the capability in one corner and starved the places it was meant to reach, while the people closest to the work, who can see exactly where it would help, sit on their hands.

It still has to be governed, of course, and the governance belongs where the work happens. A capability used everywhere has to be governed everywhere: clear decision rights about what each person can decide alone and what comes up for review, guardrails so speed does not become risk, and a shared standard so the work one person builds can be trusted by the rest. This is distributed ownership with a common discipline. It is harder to build than a single owner, and it is the only version that scales.

When my own function adopted AI, the value did not come from a central team handing tools down. It came from the conditions we built so people across the function could use them: the champions who taught their peers, the open office hours where people learned from each other, the thresholds that told everyone what they could build on their own and what had to come up for review. One of our most prolific builders was an investigator, not a technologist. The capability lived everywhere, not in one corner of the org chart.

The test is simple: do the people doing the actual work treat AI as part of their job, or as someone else's? When every good idea has to route through one group, you have turned a capability that should belong to everyone into something only a few people ever touch. When the ideas come from everywhere and the guardrails are in place, the capability is finally reaching the places it was built for.

A spreadsheet never needed an owner to change the world. It needed to be in everyone's hands, with a few rules about how to use it well. AI is the same, on a scale we have never seen. Put it in everyone's hands. Build the discipline in. Do not stand up a department and call it done.