The industry is converging on filesystems for agents

If you search for “filesystem agents” in 2026, you will find more than blog posts about coding assistants. You will find infrastructure theses, hands-on courses, and enterprise context layers all pointing at the same abstraction: give agents a navigable filesystem.

Three pieces are especially useful right now:

1. Amplify Partners — File systems for agents
   Why agents prefer FS-like interfaces over OLTP databases and object stores as the primary way they touch data — and why today’s POSIX stacks may still need to evolve for agent-scale concurrency.

2. Vercel Academy — Building Filesystem Agents
   A practical course: structure domain data as files, give the agent bash in a sandbox, keep context minimal and debuggable.

3. Box — Filesystems as the context layer
   The important production twist: a filesystem is an interface contract. Backends can be Box, SQLite-generated virtual files, or local disk — the agent still just ls / read / write.

Where FILESYSTEM.md fits

Those articles mostly answer: why filesystems? and how do I run an agent against files?

FILESYSTEM.md answers a different question: inside a project workspace, what should every agent be able to find?

• Deterministic bootstrap
• Shared behavior via AGENTS.md (with thin harness adapters)
• Layered memory, skills, and append-only logs
• Portable rules across Codex, Cursor, Claude Code, and friends

In other words:

Layer	Examples
Storage / backend FS	Local disk, Vercel Sandbox, Box, composite Deep Agents backends
Agent tools	bash, read_file, write_file, grep
Workspace contract	FILESYSTEM.md + AGENTS.md

We wrote a longer orientation page that cites these sources and maps the stack:

→ Why Filesystems for Agents

If you only have five minutes: read that page, then the Manifesto, then Getting Started.

The curated list (including Amplify / Vercel Academy / Box) lives on the Reading List.