New AI 'Dreaming' Feature Fuels Enterprise Adoption

Anthropic has introduced three major updates to its Claude Managed Agents platform, enhancing AI agent capabilities for enterprises. The new 'dreaming' feature allows agents to autonomously review past sessions and memory stores, identifying patterns and curating learnings to improve future performance without altering underlying model weights. This scheduled process helps agents learn from recurring mistakes and shared team preferences.

Two previously experimental features, outcomes and multi-agent orchestration, are now available in public beta. Outcomes enable agents to iterate towards user-defined success rubrics, with a separate grader agent ensuring adherence to criteria. Multi-agent orchestration allows a lead agent to decompose complex tasks, delegating subtasks to specialist agents, each with its own context window, improving efficiency on intricate workflows.

These advancements address key enterprise challenges: accuracy, learning, and scalability. Early adopters have reported significant improvements, such as a six-fold increase in task completion rates for legal AI company Harvey. The company also disclosed extraordinary growth, with an 80x annualized increase in revenue and usage in the first quarter of 2026, significantly surpassing its own projections and leading to compute capacity challenges.

The new features aim to close the gap between AI's potential and its real-world application. Anthropic's platform push emphasizes production reliability alongside intelligence. The 'dreaming' feature, in particular, offers a unique approach to continuous improvement, a critical factor for enterprises before entrusting AI with high-stakes tasks. The company is also expanding compute availability through a partnership with SpaceX.