When More Agents Mean More Management
An agent workforce can automate a great deal of activity while still losing its cost advantage to software that delivers the same outcome with less supervision.
Developed from a conversation between Pete Winn, Andy David and Jarrad Grigg
Jarrad found the limit while experimenting with Paperclip, where a CEO agent had hired a group of other agents. Each agent pursued its own work, consumed token budget and required him to keep checking its progress. Adding agents increased output, but it also increased the amount of management he had to perform.
That burden makes agent count a poor measure of operating leverage. Pete compared a business that needs 200 agents with one that once needed 200 employees. Both may produce the intended result, but both carry a large operating layer. Autonomy does not remove the need to understand what is happening, control costs and intervene when work drifts.
The stronger competitor is the business that turns repeatable work into software. If five people supported by purpose-built software can deliver the same result as 200 agents, the agent-heavy company can be undercut even if every task is technically automated. Agents remain useful for decisions and transformations that need judgement, but stable steps can collapse into scripts and other deterministic software. The end game is not the largest digital workforce. It is the smallest reliable operating system for the outcome.