Claude Code Auto Mode, Full Computer Use, and the Agentic Arms Race — Zvi's Issue #6
Zvi Mowshowitz has published his sixth edition of the Claude Code + Cowork + Codex roundup, and it's the most thorough independent synthesis of this week's announcements yet. Three things are happening at once, and Zvi's piece is the best place to understand why they matter together rather than in isolation.
Auto mode is the centerpiece: Claude Code now decides for itself which tool calls are safe to execute without asking you, with an AI safety classifier as the backstop blocking destructive actions like mass file deletions or credential exfiltration. Full computer use in Claude Cowork means that when a direct connector doesn't exist, Claude falls back to keyboard and mouse — anything you can do on a desktop, Claude can do for you. And Dispatch plus Channels lets you assign tasks from your iPhone or a messaging app and return to a finished result.
Zvi's framing cuts to the core of what's actually happening: Anthropic isn't checking feature boxes. It's methodically closing the gap between "AI that helps you code" and "AI that is an employee who can use a computer." The piece benchmarks this against GitHub Copilot Workspace, OpenAI Codex, and Cursor, and asks the honest question — at what point does a background worker with full computer access stop being a novelty and become the default development environment?
If you read one piece today on the state of agentic coding tools and the competitive dynamics driving them, make it this one.