For the first two months we charged per seat, because that is what every pricing page we had ever read did. It took one support thread to see the problem.
Seats punish the wrong person
A customer wrote in asking whether they should remove two colleagues from the workspace. They were not using the agent much. They were the two people who approved its riskiest actions.
Per-seat pricing had made the safety mechanism a line item.
What we changed
A run is one goal, from the first plan to either a result or a rejected handoff. Retries inside a run are free. Approvers are free. A run that stalls past its step budget is cancelled and not billed.
- If it does not finish, you do not pay
- If you decline the handoff, you do not pay
- If it retries a tool six times and then succeeds, that is still one run
The part we got wrong first
Our initial version billed on tool calls. It was simpler to implement and completely wrong: the agent could do the same work in three calls or thirty depending on the day, and customers had no way to predict a bill.
Charging for outcomes rather than effort moved the variance onto us, which is where it belongs — we are the ones who can reduce it.