Claude Managed Agents Now Remember: Anthropic Expands Enterprise AI Capabilities

Anthropic has introduced memory for Claude Managed Agents in public beta, adding a persistent storage layer that allows AI agents to retain information across sessions and improve performance over time.

The new capability lets agents store and access memories as files mounted directly to a filesystem, enabling them to build on past interactions without requiring manual prompt updates. Anthropic says this approach is designed to help agents “learn from every session,” while remaining compatible with the same tools—like code execution and bash environments—that power existing agent workflows.

The feature is aimed squarely at enterprise use cases. Memory stores can be shared across multiple agents with defined permission levels, such as read-only organizational knowledge bases or writable user-specific stores. Developers can manage these memories programmatically through an API, export them as needed, and maintain visibility into how data is created and modified.

Anthropic is emphasizing control and auditability as core parts of the release. Each memory change is logged with details about which agent and session generated it, and teams can roll back or edit stored information. These updates also appear in the Claude Console, giving developers a way to trace how agents evolve over time.

Early adopters are already using the feature to reduce repetitive work and improve accuracy. Anthropic said Netflix is using memory to carry forward insights and corrections across sessions, while Rakuten has applied it to long-running agents that learn from prior mistakes, cutting first-pass errors by 97%. Other users, including Wisedocs and Ando, are using shared memory to streamline document verification and workplace communication workflows.

The release reflects a broader push to make AI agents more useful in production environments. Instead of relying on stateless interactions, agents can now accumulate context and refine their behavior, particularly in workflows that span multiple sessions or require coordination across systems.

Anthropic positions the update as a way to extend the capabilities of Managed Agents without adding separate infrastructure. By embedding memory directly into the agent environment, the company is targeting developers who need persistent, auditable systems that can operate reliably at scale.

Memory for Claude Managed Agents is available now in public beta, as Anthropic continues to expand its platform for building and deploying enterprise-grade AI agents.

This analysis is based on reporting from Claude.

Image courtesy of Claude.

This article was generated with AI assistance and reviewed for accuracy and quality.