Every week someone asks for ten agents. We ship them one. They never ask for the other nine.
The fleet ask is a category artifact, not a customer requirement. The category sold this picture: hundreds of agents, each on a different task, all coordinating through some manager-of-managers. It looks impressive in a slide. In practice it collapses under the cost of context.
Each agent needs context — tools, history, voice, approval rules. Ten agents need ten contexts. Ten contexts go out of sync the first week a process changes. By month two the customer is debugging the fleet’s internal politics instead of getting work done.
One well-equipped employee with the same total tool access usually outperforms a five-agent fleet on the same work. We have tested this. The reason is not technical. It is organisational. The single employee has one shared memory, one rhythm, one accountable surface. A fleet has handoffs, and every handoff is a place context drops.
The customer does not lose by buying one. They gain by avoiding the operational overhead of running ten. If volume eventually justifies more capacity, we add it the way an HR team would — a second employee with their own scope, their own name, their own approval rules. Not a swarm.
The honest version of the pitch is: we are not selling you AI. We are selling you a teammate. Teammates are people, not pluralities. The day you need ten of them is the day you have found product-market fit on your AI hires, and we will know it because the first one has been profitable for six months running.
Until then, one is enough. One is usually too many for what most teams have actually scoped.
Related Start with one workflow →