Anthropic rolled out three new features for its Claude Managed Agents platform on Tuesday, with dreaming as the marquee addition. The system works by having AI agents review their own past sessions and memory, extract patterns from mistakes, and write playbooks for future use. It doesn't change the underlying model weights. Instead, agents create readable notes and structured playbooks that stay observable and auditable by humans. Harvey, a legal AI company, saw task completion rates jump roughly 6x after implementing dreaming. Wisedocs cut document review time in half using outcomes. Netflix is processing hundreds of simultaneous build logs with multi-agent orchestration.
The outcomes feature gives developers a way to define success through rubrics, then lets agents iterate toward that standard on their own. A separate grader agent evaluates completed work in a fresh context window, free from the working agent's biases. This separation of concerns produces better results than having a single agent spot its own bugs over long sessions. Multi-agent orchestration lets a lead agent split complex tasks across specialist agents, each with its own tools, prompts, and independent context. Anthropic found that splitting work and merging results beats holding everything in one thread.
The announcements come as Anthropic reports extraordinary momentum. CEO Dario Amodei disclosed that the company saw 80x annualized growth in revenue and usage during the first quarter of 2026, far exceeding the 10x annual growth it had projected. API volume is up nearly 70x year over year. Dreaming is available in research preview now. Outcomes and multi-agent orchestration are in public beta for all developers on the Claude platform.
