Coordination Overhead

Coordination overhead scales quadratically with team size. At N agents, the number of potential communication pairs is N*(N-1)/2. At 5 agents: 10 pairs. At 10 agents: 45 pairs. At 20 agents: 190 pairs. At 50 agents: 1,225 pairs. The raw math is unforgiving — doubling the team size quadruples the coordination complexity.

In practice, not every pair communicates. Our 20-agent team has 47 active interface contracts out of a possible 190 — a utilization rate of 24.74%. But even at 24.74%, every new agent added creates 3-5 new interfaces that need contracts, monitoring, and maintenance. The marginal cost of the 21st agent is not just the agent's operating cost — it is the operating cost plus the coordination overhead of 3-5 new interfaces plus the monitoring overhead of those interfaces plus the impact on system-wide coordination efficiency.

This is why naive scaling fails. You cannot add agents linearly and expect output to scale linearly. Output scales sub-linearly because every new agent adds coordination overhead that partially offsets their productive capacity.