Pods as Workers, Not Agents: Rethinking the Deployment Unit for AI Agents on Kubernetes

The kagent project proposes moving away from the traditional model of assigning one Kubernetes Pod to every AI agent, citing the inefficiency of allocating dedicated resources to bursty or idle tasks. By introducing a new control plane called agent-substrate, the project aims to treat agents as logical units that can share infrastructure more effectively. This approach seeks to better accommodate the unpredictable lifecycles of modern agents, including those that spawn sub-processes or require human oversight during execution.