Module 1 · Compute and execution · Lesson 1
Where agents run
Agents spend most of their life waiting on models and people. How that shapes the choice between Fluid functions, containers, Modal, Claude Managed Agents and eve on Vercel, with real rates and a costed fleet.
By the end of this lesson you can
- Profile an agent workload by wall time, active CPU, duration, state and burstiness
- Compare Fluid compute, containers, Modal, Claude Managed Agents and eve on Vercel on those dimensions
- Compute the compute cost of an agent run under Active CPU, per-second container and session-hour pricing, and compare it with token cost
- Split an agent into a turn handler, an execution environment and a durable waiting layer, and place each part