Anthropic Empowers Claude with ‘Dreaming’ for Self-Learning

Anthropic updates Claude with 'dreaming' for self-learning and more.

Anthropic has updated its platform for managing agents. The Claude service now features ‘dreaming’, outcome evaluation, and orchestration of multiple bots.

Dreaming enables agents to analyze past sessions and identify patterns in their operations. The system detects recurring errors and successful scenarios, updating algorithms based on these insights without human intervention. Users can set the AI’s memory to update automatically or approve changes manually.

Developers have also introduced Outcomes and Webhooks tools. Application creators can now set quality criteria for agent performance. A separate ‘evaluator’ checks results against these standards independently of the main task execution process. If the outcome does not meet the standard, the agent proceeds to correct errors.

According to Anthropic, Outcomes has increased the efficiency of complex task execution by 10%. The quality of document generation in .docx and .pptx formats has improved by an average of 9%.

The third update involves the orchestration of multiple agents. A ‘lead’ AI can delegate subtasks to specialized agents: one analyzes code deployment history, another reviews error logs, and a third handles support tickets. They all work concurrently on a shared set of files.

Harvey and Netflix Test Updates

The new features are already being tested by the company’s partners. The Harvey platform uses Dreaming for preparing legal documents, increasing task completion speed sixfold. The Netflix team has employed the multi-agent system for analyzing application build logs.

The Dreaming feature is available in preview mode upon request. Outcomes and multi-agent orchestration have entered public beta testing.

Back in May, Anthropic co-founder Jack Clark predicted the emergence of ‘self-developing AI’ by 2028.