Uploaded May 2026 | Updated September 2026, 2 weeks ago
At Carter Rabasa's session during AI Dev 26 x San Francisco, attendees learned why LLMs are already highly effective at working with file systems — thanks to decades of training on code, operating systems, and file-based workflows — and how to leverage that intuition in agent design.
The talk explored how file systems provide a powerful foundation for long-term memory and state, enabling agents to persist, organize, and revisit work far more reliably than prompt-based approaches.
It also showed how file systems act as a universal interface for data interoperability and human-in-the-loop collaboration, making them a natural layer for multi-agent and human-agent workflows.
At Carter Rabasa's session during AI Dev 26 x San Francisco, attendees learned why LLMs are already highly effective at working with file systems — thanks to decades of training on code, operating systems, and file-based workflows — and how to leverage that intuition in agent design.
The talk explored how file systems provide a powerful foundation for long-term memory and state, enabling agents to persist, organize, and revisit work far more reliably than prompt-based approaches.
It also showed how file systems act as a universal interface for data interoperability and human-in-the-loop collaboration, making them a natural layer for multi-agent and human-agent workflows.
