What I got wrong trying to build a creator team for AI brand work
The mandate was open-ended by design — a pilot project, run through internal marketing, to explore what AI could actually do for brand video work at a large corporate. No established playbook, no precedent internally. Just a brief to figure it out. I thought the challenge would be finding the right creators to experiment alongside. It turned out the challenge was something I hadn’t anticipated: I didn’t know enough about the process yet to bring anyone else into it.
That took longer to admit than it should have.
Budget conversations happened first and should have been a warning sign. Even on a pilot, stakeholders were asking for numbers, and I was building line items for a process nobody had run before. AI-assisted video work — particularly anything built around iteration and experimentation — doesn’t map cleanly onto traditional production categories. How do you estimate time spent generating outputs that don’t ship but shape the direction? How do you price a review cycle when the number of variations is essentially a creative decision?
I made guesses. The client did too. The budgets that came out were optimistic in ways that only became visible once the work started.
But the deeper problem was the brief. What I didn’t see clearly enough was that a conventional brief works because it lands inside a shared understanding of process. What I was trying to do was bring creators into a workflow that wasn’t fully designed yet — brief people on something I was still figuring out myself. On a pilot project, where the whole point is to learn, that gap is even harder to close. There’s no reference point to hand someone. No “here’s how we usually do this.”
So the onboarding kept getting deferred. There was always a reason to wait until things were clearer. In the meantime I was working directly with one AI studio, learning by being inside the work. Projects moved. Things got delivered. But the broader creator team I’d planned to build never formed.
Working closely with one studio meant I was learning, but narrowly. The workflow got built around just the two of us, and it became harder to imagine how someone new would enter it. The scaffolding that would have made onboarding possible was never built because there was no one to build it for.
What I understand now is that the onboarding problem couldn’t be solved after the workflow existed. It had to be part of designing it — especially on a pilot, where the process itself is the deliverable as much as the work. That means mapping what you know and naming what you don’t before anyone else comes in. It means being honest with stakeholders that a budget for experimentation looks different from a production budget, and that optimistic line items will collapse under the actual shape of the work.
A lot of people running AI pilots inside organisations right now are in a version of this situation — trying to staff and structure something whose process they’re still discovering. The instinct to wait until things are clearer before bringing others in is understandable. But it tends to produce the outcome I ended up with: doing the work more alone than you intended, and building something only one or two people really understand.
Tenny Tomas 