Claude is becoming less like a chatbot and more like an AI coworker that can actually get things done.
Anthropic has expanded computer use in Claude Cowork and Claude Code, allowing Claude to interact directly with your computer. Instead of simply telling you what to do, Claude can click buttons, type text, open applications, use your browser and work with desktop tools much like a person sitting in front of your PC.
That could make Claude considerably more useful for long-running tasks where simply generating text or code isn’t enough.
Claude Can Now Use Your Computer
The new capability allows Claude to interact with the desktop when it doesn’t have a direct connector or tool for completing a task.
Claude can navigate the screen, click on interface elements, type information, open files and applications, use a browser and interact with development tools.
In other words, instead of saying:
“Here’s how you can do this.”
Claude can increasingly take the approach of:
“I’ll do it for you.”
This is particularly interesting in Claude Cowork, which is designed for delegating longer, multi-step knowledge-work tasks, as well as Claude Code, where developers can give Claude more hands-on control over their development environment.
You Can Keep Working While Claude Works
One of the biggest advantages is the ability to hand Claude a task and then focus on something else.
For example, you could ask Claude to work through a spreadsheet, investigate information across multiple sources, prepare a report or interact with an application on your computer.
Anthropic says Claude can continue working while you step away, provided the computer remains awake and the Claude Desktop application stays open for computer-use tasks.
This changes the way an AI assistant can fit into a workflow.
Instead of constantly watching the chat window and responding to every small instruction, users can increasingly delegate an entire objective and let Claude work through the steps.
Claude Cowork Gets Much More Powerful
Claude Cowork was already designed around the idea of delegating work rather than simply asking questions.
It can work with local files, connected services and multi-step tasks. Anthropic describes Cowork as a way to hand off knowledge-work projects and receive completed deliverables instead of individual AI-generated responses.
Computer use adds another layer.
If an application doesn’t have a dedicated Claude connector, Claude can potentially interact with it through the screen instead.
That means Claude isn’t limited to the services that have official integrations.
For example, Anthropic highlights workflows such as:
- Filling and formatting spreadsheets
- Working with local files
- Navigating internal dashboards
- Using applications without dedicated connectors
- Running development tools
- Testing software through an app interface
- Interacting with a phone simulator
- Completing multi-step desktop workflows
The important distinction is that Claude isn’t merely analyzing screenshots. It can use the interface to actually perform actions.
Claude Code Can Also Take Control
Developers may find the feature even more interesting in Claude Code.
Claude Code can already write code, modify projects and run development workflows. With computer use, Claude can also interact with applications and interfaces on the machine when a task requires more than terminal-based work.
That could be useful for testing an application, opening an IDE, interacting with a simulator or checking how a user interface behaves.
Anthropic has previously described computer use in Claude Code as a way for Claude to point, click and navigate the screen when the necessary tools aren’t otherwise available.
This moves coding agents another step toward becoming autonomous software-development assistants.
It Isn’t Completely Hands-Off Yet
There is an important catch.
Computer use is currently a research preview, and Anthropic warns that it won’t always work perfectly. Complex workflows can require another attempt, while interacting with a screen can be slower and less reliable than using a dedicated integration.
Claude also needs access to the applications it is going to use.
Anthropic has built permission and safety mechanisms around computer use, but the company acknowledges that the safeguards aren’t perfect. Claude can make mistakes, and users should monitor what it is doing, particularly when tasks involve real-world consequences.
There Are Serious Privacy Considerations
Giving an AI control over your desktop also creates a completely different category of risk.
Claude uses screenshots to understand what’s happening on your screen. That means information visible on the desktop can potentially be seen by Claude, including private documents or other sensitive information.
Anthropic recommends being particularly careful with applications containing financial, healthcare or other sensitive personal information. Users can also block certain applications from being accessed.
This is an important difference between asking an AI to generate text and giving it control of a computer.
A wrong answer can be ignored.
A wrong click can potentially change something.
How to Try Claude Computer Use
Computer use is available as a research preview for Claude Pro and Claude Max users through the Claude Desktop application on Windows and macOS.
To get started, Anthropic says users should:
- Install or update the latest Claude Desktop app.
- Open Settings > General.
- Enable the Computer use option.
- Start Claude Cowork or Claude Code.
- Give Claude a task that requires interacting with an application.
- Approve application access when prompted.
The computer needs to remain awake, and the Claude Desktop app needs to stay open for Claude to interact with the local machine.
The Bigger Picture
The most interesting part of this update isn’t that Claude can click a button.
It’s that AI assistants are increasingly moving from answering questions to performing tasks.
We’ve already seen AI agents gain access to browsers, files, email, calendars and development environments. Giving Claude direct control of desktop applications removes another layer between an AI instruction and the action required to complete it.
That could eventually make traditional “open this app, click here, copy this information, paste it there and save the file” workflows feel increasingly outdated.
For now, Claude’s computer-use capability is still experimental, and users should treat it accordingly. But the direction is clear: Anthropic wants Claude to be able to work alongside you, not just talk to you.
For continuing updates on OpenAI, ChatGPT, Codex, GPT models, and AI industry developments, bookmark:
