Anthropic has released an updated version of its agent management platform. Claude has added the "dreaming" mode: the neurose reviews previous work sessions, identifies common mistakes and successful solutions, and then adjusts its own algorithms based on this. The user can enable automatic AI memory refresh or leave it to them to make manual adjustments.

Other innovations include the Outcomes and Webhooks tools, which can be used to define quality criteria for an agent. The special inspection module evaluates the result according to the rules - regardless of how the task was performed. If the response does not pass, the agent takes over the correction. According to Anthropic statistics, complex tasks have become 10% more efficient and the generation of documents in the formats . docx and . pptx rose as an average of 9%. The third innovation is multi-agent coordination: the main AI distributes sub-tasks among highly specialized bots.

Partners have already tried out new features. Harvey’s platform with Dreaming has accelerated document preparation six times. Netflix uses a multi-agent system to parse the application’s build logs. Dreaming mode is still available on request in preview mode, and Outcomes and multi-agent orchestration have moved to public beta testing.