Ten yes/no questions. They take a minute to answer and give you a useful, opinionated read on whether an agent fits the work you have in mind — or whether the work itself needs to change first.
Each question maps to a real cause of agent failure we've seen on engagements. The score is a guide, not a verdict; the body of the recommendation is where the value is.
So you can argue with the framing — or use it to teach a team.
Agents fail in a small number of repeatable ways. They fail when the input isn't structured enough for them to act on it. They fail when the edge cases aren't enumerable. They fail when nobody can describe how the senior person does it today, so the agent has nothing to learn from. And they fail loudest when there's no way to verify the output — which is when small errors accumulate quietly into a real problem.
The ten questions above are the inverse of those failure modes. A process that scores well on them is one where the agent has a fighting chance. A process that scores badly is one where the honest answer is fix the workflow first — the AI part is almost always easier than the workflow part.
More on this in the journal: When AI agents make sense (and when they don't) →
A short brief is enough to start. We read every enquiry ourselves and reply within two working days.