Anthropic is gearing up for a significant revamp of its Claude Code desktop experience, internally codenamed "Epitaxy." This project first emerged in late March alongside a broader source code leak. While initial glimpses suggested an experimental approach with playful animations and unclear direction, new developments indicate Epitaxy is maturing into a robust interface for power users.
The updated design takes inspiration from Cowork, Anthropic's desktop automation environment. It introduces dedicated panels for viewing a Plan, tracking Tasks executed by sub-agents, and reviewing Diffs—all within a single window. Users will also gain the ability to preview running code directly in the app and work across multiple repositories simultaneously, addressing a key limitation of the current single-session workflow.
Coordinator Mode: Orchestrating Sub-Agents
Alongside the UI refresh, Anthropic is developing a Coordinator Mode. This feature would enable Claude to act as an orchestrator, delegating implementation work across parallel sub-agents while focusing on planning and synthesis. Claude Code already supports sub-agents and experimental agent teams in the command-line interface (CLI), but Coordinator Mode appears designed to bring that capability into the desktop app with a more structured, visual interface.
The ability to create custom agents directly from within the app adds another layer, potentially lowering the barrier for non-CLI users who want to define specialized workflows. The framing, "delegate work across parallel agents," draws a natural comparison to OpenAI's Codex, which features a "Magic TODO" task system where agents work through items in parallel chats. Coordinator Mode may represent Anthropic's answer to this pattern, but implemented locally rather than in the cloud, aligning with the company's desktop-first approach to agentic coding.
https://twitter.com/ajambrosino/status/2043415431667741090
Both companies appear poised to ship desktop updates as early as next week, intensifying a race where the unit of competition has shifted from model benchmarks to developer workflow integration.