Vibe Check: Claude Cowork Is Claude Code for the Rest of Us — Katie Parrott

Katie Parrott reviews Claude Cowork, a new tab in Anthropic’s desktop app that brings Claude Code’s agentic, asynchronous workflow to non-developers. Built in roughly a week and a half, Cowork runs locally on macOS, can read/edit approved files, and browse Chrome sessions including logged-in apps.

Key distinction from chat: Cowork allows queuing messages while it works (like leaving notes for a colleague), and tasks run 30-60+ minutes without requiring continuation prompts. Skills auto-load into the Cowork environment. Available exclusively on macOS for Claude Max subscribers ($100 tier) as a research preview.

Reviewer consensus: concept is genuinely novel (no competitor offers a GUI-wrapped agentic assistant with local computer access for non-developers), but execution has rough edges. Anthropic is staking out the “asynchronous AI assistant” category before anyone else defines it. Dan Shipper notes the learning curve is real — non-technical users aren’t trained to think about async AI work.

RDCO Mapping

Directly relevant to our own agent architecture. Cowork validates the model we’ve built with the channels agent on the Mac Mini — an always-on, async AI that works through tasks without requiring real-time supervision. Our approach via Claude Code + tmux + LaunchAgent is the developer-native version of what Cowork packages for general users. The Skills integration mirrors our ~/.claude/skills/ pattern — the same concept of packaged instructions that customize agent behavior. Worth tracking Cowork’s evolution as a potential alternative interface for non-technical RDCO stakeholders.

See also: arscontexta vault agent series for related skill-graph architecture patterns.