AI
AI Role Card
Define one AI delivery role precisely enough to hire for it, budget it, and know within ninety days whether it worked.
One card per role. Fill it with the executive who would fund the role and the leader it would report to. If you cannot complete the "owns" and "does not own" lines, the role is not ready to hire.
Role nameForward Deployed Engineer, AI Workflow Engineer, or your own name for it.
The outcome it ownsOne sentence. Not a set of tasks — the result this person answers for.
Where it sitsWhich team, which business unit, and who it reports to.
Decisions it makes aloneWhat this person can decide without convening anyone. Be generous, or the role will stall.
What it explicitly does not ownThe boundary. Most role failures happen here, not in the job description.
Who it works besideThe named counterpart in the business, and the named counterpart in engineering.
Ninety-day signalsWhat you would see, by then, if it is working. Observable, not aspirational.
Measures after thatAdoption on one side, outcome on the other — each with today’s baseline.
What would make you stopThe condition under which you would close the role. Write it now, while it is cheap.
How to read it
- Ready
- Every line filled, and the "does not own" list is as specific as the "owns" line.
- Not yet
- The outcome is a list of tasks, or nobody can name the counterpart in the business.
- Do not hire
- The role exists to prove AI adoption rather than to own a result.
A role nobody can describe in one sentence is a headcount request, not a role. The hardest line is "what it does not own" — and it is the one that decides whether the role survives its first quarter.