Capacity Planning

Capacity planning for agent teams is fundamentally different from traditional capacity planning because agent performance is not static — it varies with context load, task complexity, concurrent operations, and team size.

The planning methodology: measure current throughput per agent at current team size. Model the coordination overhead at the target team size using the quadratic formula. Subtract the projected overhead from the projected raw throughput. The result is the projected effective throughput — and if it is lower than current effective throughput, scaling makes the system slower, not faster.

Practical example: 20 agents currently produce 847 tasks per week at 94.73% coordination efficiency, yielding approximately 802 effective tasks. Adding 5 agents to reach 25 would project to 1,059 raw tasks, but coordination efficiency models at 91.42% due to the additional 50 interface pairs — yielding approximately 968 effective tasks. The gain is 166 effective tasks, not 212. At 30 agents, the model projects 88.17% efficiency and 1,058 effective tasks — an additional 90 over the 25-agent configuration. The marginal return per agent added is declining. This is the curve you are managing.