Claude Code, Cowork and Codex #6: Claude Code Auto Use and Full Cowork Computer Use

Anthropic has rolled out substantial upgrades to its Claude Code and Claude Cowork AI agents, pushing them closer to autonomous operation. The headline feature is Claude Code's new 'auto mode,' which uses a classifier to review and automatically approve safe file writes and bash commands without requiring user permission for every action. Only potentially destructive actions trigger permission prompts. Currently available as a research preview for Claude Team plans (with Enterprise and API access rolling out soon), users enable it with `claude --enable-auto-mode` and toggle with Shift+Tab. Anthropic recommends using it in isolated environments, though early adopters are already running it on primary systems.

Simultaneously, Claude Cowork has gained the ability to fully control a user's computer, operating the mouse, keyboard, and screen to use any application. This 'full computer use' capability is designed to work with Dispatch, allowing users to remotely command Claude on their desktop from their phone via channels like Telegram or Discord. This creates a powerful but potentially risky remote agent scenario. The combination of autonomous coding decisions and full system control marks a major evolution in AI agents moving from assistants to active, independent operators on user systems.