Anthropic has officially launched three major enhancements to its Claude Managed Agents platform this month, further lowering the barrier to building and deploying cloud-based AI agents. First introduced just last month, Claude Managed Agents already simplified complex development workflows. This update focuses on memory enhancement, outcome definition, and multi-agent collaboration, giving the system a teamwork capability closer to that of a human team when handling complex tasks.
Among the three new features, the most eye-catching is Dreaming. According to Anthropic, Dreaming is a periodic "review" process that automatically analyzes the agent's past conversation logs and stored memories, extracts behavioral patterns, and optimizes memory storage — enabling the agent to self-improve and consistently boost performance over time. Meanwhile, the newly introduced Outcomes feature allows users to clearly define what a "successful result" looks like, and the agent will plan and execute tasks accordingly. Finally, the Multiagent Orchestration tool empowers a "lead agent" to break down large tasks into subtasks, delegating each to specialized sub-agents that have their own models, prompts, and toolkits — enabling efficient distributed collaboration.
