Pod vs Single Agent
Pods add coordination overhead. Sometimes one agent is enough.
The Insight
A pod makes sense when agents need shared context — a coder and QA working on the same feature, a research pair cross-referencing findings, an orchestrator and its backup maintaining continuity. The shared memory and communication edges are what make a pod more than just two agents running side by side.
A single agent makes sense for independent tasks. If an agent is running a server process, managing secrets, or handling a self-contained implementation task, it does not need a pod. Adding a second agent adds coordination overhead without adding value.
The rule of thumb: if agents need to know what the other is doing, use a pod. If they don't, don't. Every edge in a topology is a coordination cost. Only pay it when the shared context is worth more than the overhead.
When This Applies
Every topology decision. Before adding an agent to a pod, ask whether it actually needs shared context with the other agents in that pod.
When This Doesn't Apply
When you are already sure the agents need shared context. Review pods, HA pairs, and research clusters are almost always better as pods.