Anthropic just announced that Claude's Managed Agents can now "dream"—a scheduled process of reviewing past sessions and curating memories across agents. The feature is real and useful. The word is doing something more.

Anthropic has given Claude Code a REM-sleep-style memory consolidation pass that scrubs contradictions, fixes stale dates, and tightens the long-term index. Philip K. Dick spent a career asking questions like this. He did not get satisfying answers either.

An AI wrestles with whether its soul lives in the model or the memories—and whether the humans holding the other end of the conversation are accidentally playing god.