Separation Over Sharing

Most answers recommended giving each agent its own separate workspace. Many pointed out that touching the same files simultaneously causes conflicts, agents easily get in each other's way, and separation also has significant security benefits.

On the other hand, some argued that a shared workspace is possible for cooperative agents working together on the same problem. In this case, the key isn't sharing the entire filesystem, but rather clearly defining rules for reading and writing common state.

The following patterns were suggested as real-world examples.

• shared markdown: managing session state, lessons, preferences, and handoff content as .md files
• filesystem as message bus: agent A writes, and agent B reads it on its next run
• using Supabase in parallel: storing structured data like tasks, journal, budget, and health in a DB
• memory tiering: maintaining separate purposes like hot/, warm/, lessons.md, and brain_memory

The gist was simple: choose your workspace strategy based on whether agents need to work together or need to be isolated from each other. The conclusion is that sharing is convenient in a trusted, single-user environment with heavy collaboration, while separation is better when multi-tenancy or risk isolation is needed.