Are we using too many agents to do our work? Someone sent me a post about The My
By Brett Bohannon · November 3, 2026 · Curated by George's Blog
Are we using too many agents to do our work?
Someone sent me a post about The Mythical Man-Month and it stuck with me. Fred Brooks wrote it in 1975. His law: adding people to a late software project makes it later. His line: the bearing of a child takes nine months, no matter how many women are assigned.
The math is communication lines, n(n-1)/2. 3 people is 3 lines. 10 is 45. 30 is 435. Agents hand off context the same way.
Google Research tested 180 agent setups (Dec 2025). On step-by-step tasks, every multi-agent setup made performance worse, by 39-70%. On tasks that split cleanly into parallel pieces, a team with one coordinator improved results by about 81%. Agents working independently without a coordinator amplified errors about 17x. With a central coordinator it was about 4x.
This is basically how I run things. One agent per area: ads is one agent, listing and catalog is another. Then a chief of staff that coordinates, checks the work, and can step in on any task for a client. A lean team.
When did it become cool to have 30+ agents? Does putting 3 agents on one workflow make it smarter or faster? I'm starting to think the opposite. Add an agent when the work splits into independent pieces, not because more feels more powerful.
How many agents are you running, and how many of them actually need to talk to each other?
Couple notes to leave you with:
1. Yes Google research was in 2025 - it’s kinda ancient now
2. Does the amount of agents correlate with less domain experience?
3. Do you count the sub agents that the frontier models split off from the tasks?