You are currently viewing Anthropic Launches Claude Cowork Feature for Mac Control and Task Automation
Featured image for Anthropic Launches Claude Cowork Feature for Mac Control and Task Automation

Anthropic Launches Claude Cowork Feature for Mac Control and Task Automation

  • Post author:
  • Post category:News
  • Post comments:0 Comments
Featured image for Anthropic Launches Claude Cowork Feature for Mac Control and Task Automation
Featured image for Anthropic Launches Claude Cowork Feature for Mac Control and Task Automation
Image sourced from theverge.com
Image sourced from theverge.com

According to CNBC, Anthropic launched Cowork, a new feature for its Claude AI that lets the model control macOS computers on its own. It can click, type, scroll, open apps, and finish tasks directly on your screen.

The Verge reports that Cowork works inside Claude’s Code and Cowork tools, handling things like opening files, running dev tools, or using browsers without any setup. It builds on computer use abilities first added to Claude 3.5 Sonnet last year.

How it works

Claude tries app connectors first, like those for Slack or Google Workspace. If none fit, it takes over the mouse, keyboard, and screen. Users approve each app upfront and can stop it anytime. Unite.AI notes it pairs with a new Dispatch tool: send tasks from your iPhone to the Mac app, where everything runs locally.

  • Starts with integrations when possible
  • Falls back to screen control for browsers or apps without APIs
  • Needs the Claude desktop app running on macOS, linked to the mobile app
  • Your Mac stays awake—no sleep mode

Who gets it

Right now, it’s a research preview for Pro subscribers at $20 a month and Max at $100 a month. macOS only, with Windows planned later. News9live points out safety steps: Claude sees screen content during use, so it warns about visible data like files or emails. Some apps block by default. VentureBeat says the move raises questions about automation, reliability, and security.

Tech behind it

Unite.AI ties the feature to Anthropic’s February 2026 buyout of Vercept AI, whose team shipped Dispatch in weeks. On the OSWorld benchmark, Claude Sonnet 4.6 hits 72.5%—close to human levels of 70-75%.

Anthropic expects hiccups on tough tasks and slower speeds versus direct APIs, but they released it early to test real use.

More stories at letsjustdoai.com

Seb

I love AI and automations, I enjoy seeing how it can make my life easier. I have a background in computational sciences and worked in academia, industry and as consultant. This is my journey about how I learn and use AI.

Leave a Reply