**Anthropic Improves Claude Managed Agents with New Functionalities**
Anthropic has recently rolled out important enhancements to its Claude Managed Agents, designed to streamline the implementation and oversight of AI agents in cloud settings. These upgrades consist of three primary features: dreaming, outcomes, and multiagent orchestration.
### Dreaming: Boosting Memory and Self-Enhancement
The inaugural feature, known as “dreaming,” is presently in a research preview stage. This capability enables Claude Managed Agents to bolster their memory functions by examining prior interactions to uncover patterns and promote self-enhancement.
Dreaming functions as a time-scheduled activity that assesses agent sessions and memory repositories, drawing out valuable insights to refine memories. Users have the option to dictate how memory modifications take place, with choices for automatic updates or manual assessments prior to implementing changes.
In conjunction with memory, dreaming establishes a holistic framework for self-enhancement. Memory captures lessons throughout agent activities, while dreaming fine-tunes and refreshes this data between sessions, ensuring agents stay relevant and proficient.
### Outcomes: Establishing Success Metrics
The second feature, outcomes, permits users to delineate what a successful result looks like for the agents. By crafting a framework that stipulates success metrics, users can steer the agents towards achieving the intended results.
An autonomous evaluator reviews the agent’s performance based on these metrics in a separate context window, guaranteeing that the evaluation is unaffected by the agent’s logic. If the output falls short of expectations, the evaluator pinpoints necessary corrections, encouraging the agent to modify its tactics. Furthermore, users can configure alerts via webhooks to notify them when a target is reached.
### Multiagent Orchestration: Joint Task Coordination
The concluding feature, multiagent orchestration, enables a leading agent to break down tasks and assign them to specialized agents, each equipped with its distinct model, prompt, and tools. This arrangement facilitates the simultaneous execution of tasks, boosting efficiency and effectiveness.
For example, a leading agent might carry out an investigation while subagents concurrently assess deployment history, error logs, metrics, and support tickets. These specialists work together within a shared filesystem, enhancing the overall context for the lead agent. The leading agent can also track progress and engage with subagents throughout the workflow, as all agents maintain memory of their actions.
### Practical Use Cases
Organizations like Netflix have already started utilizing multiagent orchestration within their platform teams, highlighting the real-world applications of these innovative features.
For additional details on Claude Managed Agents and the newest feature enhancements, check out Anthropic’s official blog.