AWS turns its S3 storage service into a file system for AI agents

Amazon Web Services is making its S3 object storage service easier for AI agents to access with the introduction of a native file system interface. The new interface, S3 Files, will eliminate a longstanding tradeoff between the low cost of S3 and the interactivity of a traditional file system or of Amazon’s Elastic File System (EFS).

“The file system presents S3 objects as files and directories, supporting all Network File System (NFS) v4.1+ operations like creating, reading, updating, and deleting files,” AWS principal developer advocate Sébastien Stormacq wrote in a blog post.

The file system can be accessed directly from any AWS compute instance, container, or function, spanning use cases from production applications to machine learning training and agentic AI systems, Stormacq said.

Analysts saw the change in accessibility as a strategic move by AWS to position S3 as a primary data layer for AI agents and modern applications, moving beyond its traditional use cases in data lakes and batch analytics.

“AWS is aligning S3 with AI, analytics, and distributed application needs where shared, low-latency file access is required on object-resident data. This addresses growing demand from machine learning training, agentic systems, and multi-node workloads that require concurrent read/write access without moving data out of S3,” said Kaustubh K, practice director at Everest Group.

Without a file system in S3, enterprises developing and deploying agentic systems and other modern applications typically had to either use a separate storage system or copy, synchronize, and stage data stored in S3, introducing latency, inconsistency, and operational overhead, said Pareekh Jain, principal analyst at Pareekh Consulting.

Some developers, said Kaustubh, turned to FUSE-based tools such as s3fs or Mountpoint to simulate file systems on top of S3, but these often lacked proper locking, consistency guarantees, and efficient update mechanisms.

In contrast, S3 Files addresses those limitations through native support for file operations, including permissions, locking, and incremental updates, Jain said.

This reduces friction for developers, he said, as they will no longer need to rewrite applications for object storage: existing file-based tools will just work. “Agents also become easier to build, as they can directly read and write files, store memory, and share data. Overall, it reduces the need for extra glue code like sync jobs, caching layers, and file adapters,” Jain said.

This also implications for CIOs, as it simplifies data architecture by bringing everything, including data lakes, file systems, and staging layers, into Amazon S3.

“This approach lowers costs by removing duplication, reducing pipelines, and cutting operational overhead, while also improving governance with a single source of truth and no scattered copies,” Jain said.

S3 Files is now generally available and can be accessed through the AWS Management Console or the Command Line Interface (CLI), where users can create, mount, and deploy file systems.