“Claude Version Lobster” is here!
Tech Today, March 24 — This morning, Anthropic released the Computer Use feature for Claude, allowing Claude to automatically control a computer to perform tasks.
In AI collaboration tools Claude Cowork and AI programming tool Claude Code, users can enable Claude to control their computers to complete tasks. Their official tweet states that anything you can do sitting in front of a computer, Claude can help you do using the computer, such as opening an app, browsing the web, filling out forms, and more.
Last week, Anthropic added a new feature called Dispatch to Claude Cowork, enabling users to operate Claude on their computers via mobile phones. Dispatch, combined with the Computer Use feature, allows users to remotely command Claude to work automatically via their phone even when not in front of a computer.
Anthropic’s official blog mentioned that they pre-released this feature to understand its use cases and limitations, just as they did when launching Claude Cowork initially.
However, the Claude Computer Use feature is an early research preview and has not been fully opened to users. Currently, only Claude Pro and Max subscription users can access it, and it only supports macOS.
Anthropic’s Claude Cowork technical lead Felix Rieseberg revealed that a Windows version of the Computer Use feature will follow in the coming weeks, and noted that Claude is currently “slow to act, much slower than humans now.”
The Anthropic blog states that users need to ensure their desktop client is awake and running, then pair it with the Claude mobile app to try transferring tasks from their phone.
Below the comments on Claude, many users lamented, “Anthropic just killed OpenClaw.”
Some users also expressed regret for those who purchased Mac minis to deploy OpenClaw or spent thousands of dollars on OpenClaw wrappers.
One user joked, “Do you remember ClawdBot from before?” Since Clawd sounds similar to Claude, Anthropic accused it of infringement and demanded a name change.
One sentence to summon Claude to work, and it can also help you debug and optimize app development.
When executing tasks, Claude first carefully selects tools and calls services like Slack and Google Calendar via connectors. When no connector is available, Claude can directly control the user’s browser, mouse, keyboard, and screen to complete tasks. It will scroll, click, open, and explore, but importantly, Claude will always seek user permission before performing these steps.
Additionally, when Claude cannot access the required tools, it will click and navigate on your screen to autonomously complete tasks. It can automatically open files, use browsers, run development tools, all without prior setup. Rieseberg mentioned that Claude can also assist users in debugging native applications under development.
For example, if you’re running late for an appointment, you can summon Claude on your phone. Prompt: I’m going to be late for my date. Can you export my presentation as a PDF and attach it to my 2 PM meeting invite?
Claude will automatically operate the computer to export the PDF and then add it to the meeting invite attachment. After completing the task, Claude will reply, “Using the version edited at 9:47 this morning, exported in highest quality.”
For instance, if you’re about to give a report but suddenly realize you’re missing materials, you can ask Claude to help automate the process. Prompt: Start the development server, take a screenshot of the library page, and send it to me before the 3 PM demo, please!
Claude will update you on its progress in real-time, such as launching applications, finding the library page, the layout, and taking screenshots.
When you’re pressed for time and can’t handle work, Claude can also lend a hand. Prompt: Can you batch process all the store photos on my desktop? Resize them to 1200 pixels PNGs and add a white logo in the lower right corner. I don’t have time to do it myself.
Claude will locate the images, create a new folder on the desktop, and place the resized, watermarked images inside.
Support mobile chat control of Claude, early preview prone to errors
Last week, Claude Cowork integrated a new feature called Dispatch, which supports users chatting with Claude and assigning tasks via mobile, with final review on the computer. Now, this feature has also been rolled out to Claude Code.
Based on Dispatch, users can have Claude automatically check emails every morning, pull fixed metrics weekly, start reports or pull requests for Claude Cowork or Claude Code sessions.
With the latest Computer Use feature layered on Dispatch, the range of applicable scenarios has expanded.
When users are out, Claude can operate the computer on their behalf. For example, while on a train, Claude can help create morning briefs, modify code in IDE, run tests, and publish PRs, advancing 3D printing projects as scheduled.
Anthropic mentioned that they have built safety measures to minimize risks.
When Claude uses the computer, its system automatically scans model activations to detect prompt injections and other activities. Users can stop Claude at any time, and Claude will always request permission before accessing new applications.
However, compared to Claude’s text and programming capabilities, the development of its computer control features is still in early stages and may make mistakes.
Therefore, Anthropic recommends that users start with their most trusted applications when trying out the new features, to avoid exposing sensitive data.
Conclusion: The era of AI computer automation is here — Anthropic’s “Lobster” is coming
This time, Anthropic introduced the Computer Use feature, enabling native control of computers within Claude Cowork and Claude Code, similar to the popular OpenClaw logic, transforming AI from simple dialogue interaction into an agent capable of directly operating desktops and autonomously executing tasks.
The difference is that Anthropic’s capability is built into its official product ecosystem as a native feature, whereas OpenClaw is a third-party open-source implementation. Unlike OpenClaw, which requires users to deploy and debug on their own, Anthropic’s new feature is ready to use out of the box—users only need to download the client to experience it.
In terms of security risks, as an open-source project, OpenClaw has high deployment barriers and obvious data and permission security risks; in contrast, Anthropic’s official solution employs permission controls and safety barriers to significantly reduce usage risks.
Currently, the Claude Computer Use feature remains in early preview, with slow performance and limited user access, far from replacing OpenClaw.
In the future, when Claude’s full computer use capabilities are launched along with its security features and ease of use, it may truly usher in the “death of OpenClaw.”
