Head of AI Mentor. For the first person in the company to hold the title.

There was no job description. There was a slide that said Head of AI with your name under it, a board that wants a plan by the next meeting, and forty tools already in use across the company that nobody inventoried. You are the first person to hold the title here. There is nobody above you who has held it either.

The mistake in month one is writing the strategy. A strategy you write alone in week two gets read as a land grab by every function head who was already doing something with AI before you arrived, and they will quietly route around it.

So the first artefact is not a strategy. It is an inventory and a map of owners: what runs today, who pays for it, what it touches, what would hurt if it broke. Then a three-tier answer for the board, because AI at your scale is three decisions with different owners: individual productivity, team workflows, product bets. The strategy comes in month three, and by then it is co-signed.

By the second board meeting: a written policy, one dashboard, and an answer to how much of this are we doing and why that holds when a director pushes on it.

The platform team reports to the CTO. The ML team reports to you. The product teams report to neither and use whatever they found on Hacker News. Every quarter the same argument: whose roadmap does the model gateway sit on, who approves a new vendor, who answers when the customer support bot says something it should not.

Most of these fights are not about AI at all. They are about an org chart that added a seat without removing anything from the seats next to it.

We write the boundary with the CTO in the room, or at least with their draft. Three columns: yours, theirs, shared with a named decider. One escalation path. The point is not winning the columns. The point is that the next fight takes twenty minutes instead of a quarter.

Adoption is at 90 percent. The engineering blog has three posts about it. The support team says tickets are faster. And the finance director has a spreadsheet showing tool spend up four times year on year and headcount unchanged.

Ninety percent of people opened the tool. That is the whole content of the adoption number, and a board member will work that out in the meeting, usually while looking at the delivery chart next to it.

The numbers I would put in front of a board instead: rework and revert rate on AI-assisted work, cycle time delta for the teams that opted in against the ones that did not, and cost per team rather than per seat. Three numbers, one page, sent monthly before anyone asks. We build it in two sessions and you own it from there.

The platform works. The policy is signed. The product teams still paste into a consumer chatbot because the internal gateway needs a ticket to get access and the ticket takes a week.

Enablement is the job most Heads of AI underestimate, because it looks like internal marketing and feels beneath the title. It is where the role is won. Microsoft's 2026 Work Trend Index across 20,000 people found that when managers model AI use themselves, their teams' trust in agentic tools rises by 30 points. Nothing in the policy doc moves that number. A manager doing the thing in front of the team does.

So we pick the three teams most likely to succeed, remove the ticket, put you in their standup for a month, and publish what changed. Then the next three. It is slower than a mandate and it is the only version that sticks.