⏱ If There's Time — The Limits of Memory

Info
⏱ Duration: ~15 min (optional block) · 🎯 Goal: understand where Claude’s memory lives —and where it does NOT— so you govern automation without assuming context that isn’t there.

When you automate, it’s easy to assume Claude “remembers everything.” It doesn’t: memory has clear boundaries, and confusing them is a silent source of errors in scheduled tasks and governed workflows.

The three boundaries

  • Chat does NOT carry into Cowork. What Claude “knows” from your chat conversations does not transfer to Cowork. A Cowork task starts without that context.
  • In Cowork, memory lives only in Projects. Within Cowork, persistent context across tasks is stored in a Project: the Project’s files, instructions, and knowledge are the shared memory. Outside a Project, each task starts cold.
  • Dispatch adds cross-session memory (Pro/Max only). Dispatch’s single thread retains context from one task to the next —cross-session memory— but this is a Pro/Max-only capability, not Team/Enterprise.
Tip
Governance implication: if your scheduled task needs context (a policy, a glossary, baseline data), don’t trust Claude to “remember it”: put it in the Project where the task runs, or attach it in the prompt. Memory that isn’t in the Project doesn’t exist for that run.

✅ Checkpoint

You can explain that chat does not share memory with Cowork, that in Cowork persistent memory lives in Projects, and that Dispatch’s cross-session memory is Pro/Max only. You know where to place the context a scheduled task needs.

Workshop Claude Avanzado · Grupo Salinas · v1.0 · 2026