Anthropic Adds Three Core Capabilities to Claude Agents

AI startup Anthropic this week announced three major updates to its cloud-hosted agent service Claude Managed Agents, aimed at enhancing autonomous agents' memory, task goal definition, and collaboration. The service officially launched last month, significantly streamlining the process of building and deploying cloud-based AI agents. The new features include Dreaming, Outcomes, and Multiagent Orchestration.

According to Anthropic, Dreaming is a scheduled execution mechanism that reviews an agent's past conversations and memory storage to extract behavior patterns and optimize memory, enabling self-improvement over time. It works alongside the existing memory system to improve agent stability across long-term tasks. The Outcomes feature allows users to specify success criteria for tasks, and the agent adjusts its execution strategy to ensure outputs meet expectations.

The most notable addition is the Multiagent Orchestration tool, which lets a "master agent" break complex work into subtasks and assign them to "expert agents" with independent models, prompts, and tools for parallel processing. This design can significantly boost efficiency and accuracy on large projects. These features are now available to Claude Managed Agents users. Industry analysts suggest Anthropic is building a competitive edge in the enterprise AI market by strengthening agent collaboration and self-evolution. In the future, such "agent clusters" could become the standard paradigm for automated workflows.