Anthropic details agent memory: "dreaming" helps Claude fix mistakes and organize knowledge
RLanceMartin · x · 2026-11-07
Anthropic's new engineering blog explains how it designs memory for Claude.
- Borrowing from neuroscience: like the hippocampus replaying experiences during sleep, agents can use idle time to reflect on logs — a process called "dreaming," inspired by Letta/UC Berkeley's sleep-time compute.
- Dreaming can correct mistakes, create notes, build new memory representations, and shape what gets preloaded into future sessions.
- The team shares several evaluations used to explore agent memory and what each component contributes to organization.
More from coding & agent
- Conductor shuts down as cloud agent startups fold, devs hunt alternatives — sergeykarayev · 2026-11-07
- Incorporate a DE C-Corp via Stripe CLI + MCP, plus $100k in Cloudflare credits — seanmcdonaldxyz · 2026-11-07
- Mohit Bansal's Northwestern lecture: fixing agent trust, world discovery, and long-horizon memory — ManlingLi_ · 2026-11-07
- Hackathon winners build GLaDOS AI escape room that phishes you for permissions — charles_irl · 2026-11-07
- Claude Duet can now access Claude Code sessions in the cloud and on connected local devices — daniel_mac8 · 2026-11-07
- Running a business on meeting transcripts: Claude Code fixes apps mid-call via Fireflies — curious_vii · 2026-11-07