Shared File Systems Are the Key Primitive for Effective Human-Agent Collaboration
The primary bottleneck for AI agent success is giving agents the right context in a form both humans and models can understand. File system interfaces serve as the ideal shared working area, enabling agents and humans to collaborate through familiar structures like plans, notes, logs, and drafts.
Human oversight of agents is easier when the agent's working memory is externalized in a human-readable shared space, making file systems a natural fit for human-in-the-loop agent workflows.
The main variable determining agent success is whether the agent can access the context it needs to do its work.
Agents using file systems is a significant development because it creates a unified system both humans and agents can use to pass around data.
A filesystem-shaped interface is more legible to both AI models and the humans supervising them than alternative data-sharing approaches.
Agents need a 'working set' โ a persistent, structured working area containing plans, notes, task lists, policies, drafts, summaries, logs, corrections, and decisions โ rather than just raw data access.
The ideal agent primitive is giving agents access to systems humans already know how to use, but optimized for how agents work โ not building entirely new paradigms.
โWhat they need is a working set: plans, notes, task lists, policies, drafts, summaries, logs, corrections, decisions, etc. For that layer, a filesystem-shaped interface tends to be more legible to both the model and the humans supervising it.โ
Shared with Earmark ยท earmark-ai.com