Box pitches 'virtual filesystem' layer for AI agents

AI-focused collaboration platform Box has a concept that AI agents need dynamic content-generating filesystems and not ones that are based on storing and retrieving static content.

The company has an AI-first content management strategy and Box CEO and co-founder Aaron Levie said in a CNBC interview: “At Box, we're building a platform that helps agents actually leverage the unstructured data, the kind of critical context in your enterprise to be able to help you go and automate your workflow. So agents actually need a file system to be able to do their work and to be able to actually go and accelerate the knowledge work that everybody's looking to bring AI into the enterprise for."

The Box idea, described in a blog, is that "a filesystem doesn't have to mean literal files on disk. It's an interface contract. The agent only needs to understand how to navigate files."

An AI agent will use a virtual filesystem with a directory tree of folders which can be interfaced to dynamically generated content. For example:

• /docs/ is backed by cloud storage
• /memories/ is backed by SQLite where relational data is synthesized into files
• /workspace/ maps to local disk for generated outputs

The agent will navigate a seemingly normal filesystem, browsing directories, carrying out file read and write operations, which are executed by backend software, translating these into actions specific to each folder item. The blog writer says: "We use a CompositeBackend to route different directory prefixes to different storage systems."

A file access path to a user profile – /memories/users/sarah-chen.json – could actually be a SQL operation with data "generated dynamically from relational tables. Rows become JSON. Table joins become readable Markdown summaries. The agent never sees SQL — it simply reads files."

This means that an agent's reasoning logic can stay the same even if the underlying operational logic changes.

The Box bloggers changed the /docs/ S3 cloud storage for a so-called BoxBackend that plugged directly into Box's Deep Agents' BackendProtocol.

Thus, "a filesystem read operation becomes a path resolution step followed by a Box file download. The agent might request something like /docs/pricing.md, but under the hood we resolve that virtual path to a Box file ID and fetch its contents via the Box API."

The agent never sees file IDs and doesn't see API calls. It just reads (what it thinks are) files. From the agent's perspective, nothing changes, as it still calls filesystem tools like ls, read_file, and write_file. The abstraction contract stays the same but what happens underneath changes, as the agent is actually interacting with a dynamic content layer.

With write operations, if the content is new, a fresh item is written, but "if a file already exists, we upload a new version instead of silently overwriting it — which means agent-generated artifacts automatically inherit version history."

The agent sees itself as simply writing a file, but the underlying Box operation enforces "folder scoping, permission-aware access, and versioned writes inside Box... interacting with a governed, collaborative, enterprise-grade content layer."

Box suggests an AI agent can draft "a [sales] proposal using governed company documentation and write the output directly into a shared workspace. It can respect existing permission models automatically. It can operate within audit and compliance boundaries."

Here’s an agent request to do this:

With this request, Box's AI agent "navigates that filesystem exactly as instructed. It lists files, opens documents, reads synthesized customer history from SQLite, drafts a proposal, and writes the output to the workspace. Because /docs/ is backed by Box, we can also upload that generated proposal back into Box with version history preserved... The agent doesn't know it's interacting with three different storage systems. It only knows it's interacting with a filesystem."

Box’s blogger thinks "filesystems are becoming a core abstraction for AI agents... A filesystem gives agents a predictable, navigable way to manage context across documents, memory, and generated outputs. It becomes a stable interface between reasoning and data — one that scales as workflows become more complex."

The filesystem is viewed as a dynamic data delivery contract instead of a static data storage mechanism, and "can be backed by relational databases, object storage, local disk, or an enterprise-grade content platform — without rewriting the agent itself."

Try Box's example here.

One quick reaction to this is that the latency and data delivery bandwidth of file read and write operations is no longer solely based on the performance of the storage media on which static file data is stored. There is intervening dynamic content generation processing taking place, which will increase latency and deliver unpredictable amounts of data.