We declared early that AI was everyone’s job. It sounded right, it went on slides, and heads nodded everywhere it was said.
Months later I noticed what the phrase had actually produced: warm agreement and thin accountability. Use cases waited for someone to sponsor them. Data issues waited for someone to own them. Learning happened where individuals were curious and didn’t where they weren’t. Everyone supported AI the way everyone supports cyber security posters. When something is everyone’s job, it is no one’s responsibility, and enthusiasm without a name attached turns out to be indistinguishable from nothing.
The correction wasn’t to centralise, which is the classic overcorrection. Handing AI to a central team, mine, for instance, would recreate the arm’s-length problem in permanent form: the business requests, the AI team delivers, nobody outside it changes how they think. We’d have built a bottleneck and called it a strategy. AI does need to live everywhere. It just can’t live everywhere anonymously.
So we made ownership specific. Every department sets its own direction for how AI changes its work, and every department has a named person accountable for that direction. Not a technologist parachuted in, one of their own leaders, close to the outcomes. Every agent has a named owner. Every use case that wants investment has a sponsor who owns the result in operations, not just the idea. And leaders answer for the capability of their own teams, because “my people haven’t been trained” is now their sentence to finish, not mine.
The central team still exists, but its job changed shape: platform, guardrails, education, and making the distributed owners successful. Enablement, not ownership.
The difference in traction iss unmistakable. Named people move. Crowds wait.
It’s worth running the test in your own organisation. Pick three things your AI ambition depends on, adoption in a specific division, the quality of a critical dataset, the readiness of a workforce, and ask who owns each one. If the answer takes more than five seconds, the answer is no one.