Anthropic Unleashes Claude as Your Digital Co-Worker — AI Can Now Control Your Mac, No Hands Required

Key Takeaways

1. Anthropic, via its official @claudeai X account, announced the feature on March 23, 2026​
2. Claude can now autonomously open apps, navigate your browser, fill spreadsheets, and complete desktop tasks. It can do anything a human would do sitting at a desk​.
3. Currently a research preview, available exclusively on macOS for Claude Pro and Max subscribers (plans starting at $20/month). It is available inside Claude Cowork and Claude Code.
4. The tweet announcing the feature racked up 14,000+ likes and 1,700+ reposts within hours, signalling massive user interest

Quick Recap

In a major step toward true AI agency, Anthropic officially announced on March 23, 2026 that users can now enable Claude to take direct control of their computers. Via its verified @claudeai account on X (formerly Twitter), the company confirmed that Claude can open applications, navigate browsers, and fill in spreadsheets. This means Claude can replicate virtually any task a human user would perform at a desk. The feature is currently rolling out as a research preview within two Anthropic products — Claude Cowork and Claude Code. Moreover, it is restricted to macOS only for now.

How Claude’s Computer Control Actually Works?

The capability is powered by what Anthropic calls a “computer use” architecture, where Claude captures screenshots of the desktop, interprets the visual state of the screen, and then issues mouse and keyboard commands to interact with it. This replicates how a human reads and responds to what’s on screen.

The system follows an observe → plan → act loop: it takes a screenshot to understand the current state, plans the next action, executes it, then verifies the result before proceeding. Critically, Claude prioritises using direct connectors for apps like Slack, Calendar, and Chrome first. If no connector exists, it only falls back to raw screen interaction. Additionally, it will ask user permission before accessing each new application.

The feature is surfaced through two distinct products. Claude Cowork is designed for non-developers — a no-code, no-terminal interface for knowledge workers who want to automate tasks like weekly report generation, batch photo editing, or data extraction from dashboards. Meanwhile, Claude Code, originally a developer-focused terminal tool launched in February 2025, extends the same agentic engine to engineering workflows. The update also introduces a “Dispatch” feature — users can assign a task from their smartphone. Then Claude completes it on the desktop while they step away.

Cross-device automation and scheduled recurring tasks (e.g., “scan my email every morning” or “pull a report every Friday”) are also now supported.

The Race to Own the Desktop: Why This Moment Matters?

This announcement lands at the peak of what analysts are calling the “agentic AI era” — a phase where AI tools are no longer reactive chatbots but proactive, task-executing agents. Anthropic is entering crowded but still-nascent territory. For example, OpenAI launched its Operator (now absorbed into “ChatGPT agent mode” as of July 2025) for users on its $200/month Pro plan. At the same time, Google has been shipping its own Gemini Computer Use capability since late 2025.

What distinguishes Anthropic’s move is accessibility. OpenAI’s comparable agent features require a $200/month subscription and remain geographically restricted to US users. Claude’s computer use is available to Pro subscribers at $20/month, dramatically lowering the barrier for individuals and teams globally. Additionally, unlike OpenAI’s Operator — which runs in a sandboxed virtual browser on OpenAI’s servers — Claude’s computer use operates directly on your local macOS environment. This gives it access to local files, apps, and tools that purely cloud-based agents cannot reach.

Anthropic has also been transparent about safety guardrails:

Claude asks permission before accessing each application, prefers integrations over raw screen control, and has classifiers designed to detect and reject prompt injection attacks — malicious instructions embedded in web pages that could hijack Claude’s behavior. The macOS-only limitation is partly a safety choice. It allows Anthropic to constrain the environment during the research preview phase before expanding to Windows.

Competitive Landscape: Claude Computer Use vs. The Field

Claude leads decisively on accessibility and local integration — its $20/month entry point is a tenth of OpenAI’s CUA price, and its ability to control local apps (not just a sandboxed cloud browser) gives it a real-world utility advantage for power users. Google Gemini Computer Use holds the edge in cost-per-token for high-volume API usage, and its deep integration across Google Workspace gives it a natural moat for enterprise teams already on Google’s ecosystem — but it currently lacks the no-code consumer interface that Cowork provides.

Sci-Tech Today’s Takeaway

In my view, this is one of the most consequential AI product moves of 2026 — and I don’t say that lightly.

I’ve been tracking the “computer use” race since Anthropic first teased the capability in October 2024, and what struck me then was the same thing that strikes me now: this is the feature that finally closes the gap between “AI as advisor” and “AI as doer.” I think this is a big deal because it democratizes automation that previously required either a developer background or a $200/month OpenAI subscription. A $20-a-month subscriber can now assign their Mac tasks by phone and come back to finished work. That’s genuinely new.

In my experience, the features that drive the biggest adoption spikes are the ones that eliminate friction for non-technical users — and Claude Cowork nailing that “no terminal, no code” angle is exactly the right instinct. The 14,000+ likes on the announcement tweet in under 24 hours tells you the market appetite is real.

My verdict? Bullish for Claude adoption, bearish for legacy task automation tools. The macOS-only limitation keeps this a preview for now, but the moment Windows support drops, this could meaningfully shift how millions of knowledge workers interact with their computers daily. Watch this space — the agentic desktop is no longer theoretical.