How do I practice shared state between agents?

The most direct way to practice shared state between agents is through a focused exercise: Identify two or three agents — cognitive routines, tools, or processes — that you run regularly and that should inform each other but currently do not. Write down what each agent produces as output and what each agent would need as input to perform better. Then design a shared state artifact: a single document, dashboard, or note that all agents read from and write to. Use it for one week. At the end of the week, note which agents made better decisions because they had access to information they previously lacked.

Common pitfall: Assuming shared state means every agent sees everything. Unrestricted shared state creates noise, not coordination. When every agent dumps its full output into a common pool, agents drown in irrelevant information and slow down. The failure is conflating access with design. Effective shared state is curated: each agent writes specific, structured outputs to defined locations, and each agent reads only the subset it needs. The opposite failure is equally destructive — no shared state at all, where agents operate in complete isolation and produce contradictory outputs because they cannot see each other's work.

This practice connects to Phase 26 (Multi-Agent Coordination) — building it as a repeatable habit compounds over time.