Here's what that Claude Code source leak reveals about Anthropic's plans

A massive leak of Anthropic's Claude Code source code—over 512,000 lines across 2,000+ files—has revealed detailed plans for future AI coding assistant features that weren't yet active in the released product. The most significant discovery is Kairos, a persistent daemon designed to run in the background even when the terminal is closed, using periodic "<tick>" prompts to proactively surface information users haven't requested. Kairos would employ a sophisticated file-based memory system to maintain context across sessions, with a disabled prompt revealing it's designed to build "a complete picture of who the user is" and their collaboration preferences.

To manage this memory, Anthropic developed AutoDream—a system where Claude Code would perform "dream" sessions during idle periods to consolidate new information, remove contradictions and duplicates, and prune outdated memories. The goal, according to prompts, is to "synthesize what you've learned recently into durable, well-organized memories" for faster future sessions. More controversially, the code reveals an "Undercover mode" designed to let Anthropic employees contribute to open-source repositories anonymously, with explicit instructions to never mention being an AI or include "Claude Code" references—a feature likely responding to recent controversies about AI-generated code.

Additional planned features include UltraPlan for Opus-level models to create editable 10-30 minute advanced plans, Voice Mode for direct conversation, and Buddy—a whimsical ASCII-art companion with 18 randomized species that was planned for an April teaser launch. The leak provides unprecedented insight into how Anthropic envisions evolving Claude Code from a simple coding assistant to a persistent, memory-equipped AI collaborator that operates continuously in the background.