Anthropic's Claude Code source leak reveals ambitious plans for AI persistence through multiple unreleased features. The leaked code shows development of a Kairos daemon designed to operate continuously in the background even when the terminal window is closed.

The Kairos system uses periodic "tick" prompts to review needed actions and includes a "PROACTIVE" flag for surfacing information users haven't requested. This persistent capability relies on a file-based memory system that maintains context across user sessions, aiming to understand user preferences and work patterns.

When users become idle or manually end a session, the AutoDream system initiates what the code calls a "dream"—a reflective pass over memory files. During this process, Claude Code scans transcripts for new information, avoids duplicates, and prunes outdated memories.

The system synthesizes recent learning into durable memories to help future sessions quickly orient themselves. These features suggest Anthropic is developing AI with near-human memory continuity that could fundamentally change how users interact with coding assistants.