ANTHROPIC TEACHES CLAUDE TO DREAM
@AnthropicAI announced new features for its agent platform on Wednesday, including a research preview of "dreaming," a background process where Claude reviews past sessions, spots recurring mistakes, and reorganizes its memory while idle.
The analogy is biological. Just as the human brain consolidates memory during sleep, dreaming agents process accumulated workflows to find patterns no single session could catch in real time. Users can run it fully automated or review changes before Claude writes them.
Two more features moved to public beta. Outcomes lets users define what "good" looks like, with a separate grader agent checking the work against those criteria. Multi-agent orchestration allows a lead agent to break down tasks and assign them to sub-agents, with a console view that shows each step.
Anthropic also doubled rate limits for Claude Code on Pro, Max, Team, and Enterprise plans, removed the peak hours reduction, and significantly raised API rate limits for Claude Opus models. All effective today.