Breaking down the components of AI agent memory and "dreaming"

Drawing on human memory, the post explains how Anthropic gives Claude effective agent memory: knowledge retained across sessions, separate from in-context learning and weight changes. It cites Letta and UC Berkeley's "sleep-time compute" and describes "dreaming"—reflecting on logs to organize memory—plus evaluations of each component.

• Fintech giant drops Claude for a Chinese model2026-11-07
• Anthropic's Claude and data center CEO to headline Rowan AI conference2026-11-06
• Study of 207,000 Claude sessions shows how scientists use AI2026-11-06
• METR study finds AI is speeding its own development2026-11-06
• Ads Arrive in AI Chatbots as Anthropic Keeps Claude Ad-Free2026-11-05