Anthropic has introduced three new agent features: Dreaming, Outcomes, and Multiagent Orchestration. Dreaming is a research preview that enhances agent memory by consolidating past sessions and identifying patterns, rather than retraining the model itself. Outcomes, in public beta, allows agents to self-evaluate their work against defined success criteria. Multiagent Orchestration, also in public beta, enables parallel execution between lead agents and sub-agents. These infrastructure-focused updates aim to improve agent deployment and performance, with early adopters like Harvey reporting significant gains in task completion rates. AI
Summary written by gemini-2.5-flash-lite from 2 sources. How we write summaries →
IMPACT These infrastructure updates for AI agents could significantly improve their reliability and performance in complex tasks.
RANK_REASON Product launch of new agent features by a major AI lab.