Claude Managed Agents Launches Dreaming Research Preview

Anthropic Unveils Multiple Agent Capabilities at Code with Claude

At the Code with Claude event, Anthropic announced a series of new feature updates for Claude Managed Agents. The centerpiece of this release is an experimental capability called Dreaming, which debuted as a research preview. Alongside it, three features — Outcomes, multiagent orchestration, and Webhooks — have entered public beta.

This flurry of announcements signals that Anthropic is doubling down on its Agent ecosystem, pushing Claude beyond a purely conversational model toward an intelligent agent platform capable of autonomously executing tasks and working collaboratively.

Dreaming: A New Concept in Research Preview

Based on the official language, Dreaming is currently in research preview — meaning it leans more toward exploratory territory than a production-ready feature. Research previews typically allow developers early access to cutting-edge capabilities and serve as feedback channels, while also signaling that the feature's interface, behavior, and overall positioning may change significantly in future iterations.

Given limited information in the announcement, Anthropic has not detailed the specific mechanics or use cases for Dreaming. Judging by the name alone, it may relate to autonomous reasoning, planning, or knowledge synthesis that an Agent performs during idle or background states — though this remains reasonable speculation pending further technical disclosure from Anthropic.

Three Features Enter Public Beta

The three capabilities entering public beta simultaneously form a more complete engineering foundation for Managed Agents:

Outcomes

The introduction of Outcomes typically means developers can more explicitly define the target states an Agent needs to achieve, enabling the agent to work toward results rather than simply executing steps. This results-oriented design helps improve the controllability and measurability of Agent behavior.

Multiagent Orchestration

Multiagent orchestration enables multiple Agents to divide labor and collaborate on complex tasks. This is one of the key directions in the current AI Agent landscape — complex workflows that a single model struggles to handle efficiently can often be completed through the coordinated efforts of multiple specialized Agents, improving overall reliability and scalability.

Webhooks

The addition of Webhooks strengthens an Agent's ability to integrate with external systems. Through Webhooks, an Agent can trigger external services when specific events occur, or receive real-time notifications from external systems — laying the groundwork for building event-driven automation workflows.

What This Means for Developers and the Industry

Moving Outcomes, multiagent orchestration, and Webhooks from internal experimentation to public beta reflects Anthropic's pragmatic approach to Agent engineering: pursue frontier exploration (Dreaming) while also building deployable infrastructure. For developers, public beta means they can validate these capabilities in real projects ahead of general availability and help shape their final form.

It's worth noting that both research previews and public betas are early-stage designations — feature stability, pricing strategies, and long-term support all carry uncertainty. Developers are advised to use these features cautiously in production environments and keep a close eye on Anthropic's official documentation updates.

Summary

The key takeaway from this Code with Claude release: Dreaming opens up exploration as a research preview, while Outcomes, multiagent orchestration, and Webhooks advance to public beta. Together, these updates push Claude Managed Agents toward a more powerful, more composable agent platform. Given the limited detail in the official announcement, more technical specifics are expected as Anthropic continues to share information.