news · ·4 min read

Claude Can Now Control Your Mac

Anthropic launched Computer Use for Mac on March 24, putting Claude in the driver's seat of your desktop — files, apps, browser, IDE, and all. Here's what changed, what it can actually do, and how it differs from OpenAI's Operator.

computer-usemacclaude-codecoworkdispatch

Claude can now take the wheel. As of March 24, 2026, Anthropic’s Computer Use for Mac is live — no waitlist, no sandbox, no browser-only prison. Claude sits on your Mac, watches your screen, and acts. Clicks, types, opens files, runs tests, submits pull requests. The full thing.

This is the first feature in the AI assistant space to give a model genuine local desktop control. It matters.

What Happened

Anthropic shipped Computer Use inside two products simultaneously: Claude Cowork (the Mac desktop app) and Claude Code (the terminal-native dev tool). Both are available now to Pro ($20/mo) and Max ($100–$200/mo) subscribers. Research preview label, but no waitlist — you can use it today.

The underlying mechanic is an observe-plan-act loop:

  1. Claude takes a screenshot
  2. Plans the next action
  3. Executes via mouse, keyboard, or native connectors
  4. Verifies the result, then repeats

It’s not magic — it’s a tight feedback loop running locally on your machine.

Claude tries to be smart about how it controls things. It follows a fallback hierarchy: native app connectors first (faster, more reliable), then a Chrome extension for browser tasks, then raw screen control as a last resort. Most of the time you won’t notice, but it means Claude isn’t just blindly pixel-hunting.

What it can actually do:

Also worth noting: Dispatch, launched a week earlier on March 17, lets you trigger Mac tasks from your iPhone via QR scan. You can queue up tasks while away from your desk — recurring ones included — and Claude handles them when your Mac is awake.

Why It Matters

The honest comparison is OpenAI’s Operator. Operator runs in a cloud-sandboxed browser. It can fill out web forms. It cannot touch your local files, your IDE, or anything outside a browser tab. Your data goes to OpenAI’s servers.

Claude’s Computer Use runs entirely on your Mac. Your files stay local. Claude has access to everything you have access to — your entire filesystem, every installed app, your dev environment, your terminal. That’s not a small difference. That’s the whole ballgame for developers.

For developers specifically: Claude Code already had /loop for cron-style background execution. Combine that with Computer Use and you have Claude running recurring local tasks autonomously — no cloud dependency, no data leaving your machine.

The community reaction has been enthusiastic, particularly on r/ClaudeCode. MacStories called it “rough around the edges but showing real promise” — which is accurate and fair.

What to Do

If you’re a developer: Update Claude Code to the latest version, then try something concrete. Give Claude a failing test suite and ask it to fix and commit. Or point it at a GitHub issue and let it take a first pass. The /loop flag in Claude Code is worth reading up on for background task scheduling.

If you’re on Cowork: Start with a well-scoped task — something like “pull the last 30 rows from this spreadsheet, reformat it, and save as CSV.” Don’t throw a 10-step workflow at it on day one. Latency is real, complex tasks sometimes need a retry, and certain apps are blocked by default.

Keep these limitations in mind:

The API-level Computer Use Tool is still in beta if you want to build on top of it directly. Expect rough edges there too.

Computer Use for Mac is the first Claude feature that makes “AI agent” feel like something real rather than a demo. The local-first architecture is the right call — it’s what separates this from browser-sandboxed cloud tools. Use it on tasks you can afford to retry. Watch what Claude does. The loop is just getting started.