Six months ago, the team named an AI champion: someone visibly good with the tools, patient enough to help, willing to take on the role. It looked like progress — a clear owner, a go-to person, momentum. Six months later, every non-trivial AI question in the team still goes through that one person. Nobody else has gotten meaningfully more confident. The champion is busier than ever, and the team’s actual AI capability hasn’t moved much beyond where it started.
Naming a champion feels like progress but isn’t capability
A designated AI champion is an easy, visible thing to point to when someone asks what a team is doing about AI adoption. But a single skilled person absorbing everyone else’s questions is not the same as the team getting better at judging AI output, framing a good request, or knowing when to trust a result and when to check it. It’s the same pattern coaching sustainability work has always had to guard against: one person becomes so central to the practice that the practice can’t survive without them.
The same anti-patterns, wearing a new title
The organizational habits that once undermined external coaching show up just as easily around an internal AI champion. Leadership delegates “the AI thing” to that person and steps back, rather than changing how the team itself reviews and owns AI-assisted work. Urgency creates a shortcut: under deadline pressure, people go straight to the champion instead of building their own judgment, because it’s faster this one time — and then the next time, and the time after. Nobody has actually defined what the champion owns versus what every team member is expected to own themselves, so the boundary quietly expands until the champion owns everything AI-related by default.
Design the role for its own obsolescence
A champion role built well starts with the opposite question from the one most teams ask. Instead of “who’s good enough to be the champion,” the useful question is “what does this team need to be able to do without the champion in six months, and how does the role get us there.” That reframes the role’s job: not to be the answer, but to close the gap between where the team is and where it needs to be able to operate on its own — the same shadow-lead-step-back arc that makes any coaching or enablement role sustainable rather than a permanent crutch.
Concretely, that means the champion’s early work looks like doing the AI-assisted task with the team watching and asking why; the middle phase looks like the team doing the task with the champion available but not driving; and the later phase looks like the team operating independently, with the champion available for genuinely hard edge cases rather than routine ones.
What actually needs to transfer is judgment, not tool tips
The habits worth transferring from a champion to a team are rarely which button to click. They’re the harder, more transferable skills: recognizing when an AI-generated answer is plausible but ungrounded, knowing what context a request needed and didn’t get, and being able to say clearly why a particular output was accepted or rejected. Prompting Is Not the Team Capability makes the same point from a different angle — the technique of asking well is not what determines whether a team can be trusted with the answer it gets back.
This is why role-based enablement on its own tends to plateau: it can transfer prompting fluency reasonably fast, but evaluation judgment only develops through the team practicing on its own real work and getting it wrong occasionally, with support close enough to catch a real mistake but not so close that no one has to decide anything themselves.
Evidence the role is working looks like less dependency, not more usage
The easy, misleading metric for an AI champion program is usage: how many people asked the champion for help, how many sessions ran, how satisfied people were. Those numbers can rise indefinitely while the underlying dependency gets worse, not better. The more honest signal is the opposite direction — whether fewer questions need the champion over time, whether the team can point to AI-assisted decisions it made and defended without the champion in the room, and whether the champion’s own role has visibly narrowed from “does everything AI-related” to “handles the genuinely hard cases.”
That is the same evidence bar Training Roles Does Not Change the Work applies to any capability-building effort: look for it in the decisions and the work, not in attendance or utilization. The Flow Cracker Playbook frames the same underlying question at the leadership level — what capability and support does a team actually need enabled, with an explicit endpoint, rather than a role that quietly becomes permanent because nobody designed it to end.
An AI champion can be a genuinely useful way to start. It stops being useful, and starts being a trap, the moment nobody notices that the team still can’t operate without them.