A generative AI company has upgraded its Claude language model so that it can directly control a user’s computer and carry out tasks on command, positioning itself as a competitor to the personal agent framework OpenClaw.
This update was announced in a blog post published on March 23, where the company introduced new capabilities within its Claude Code and Claude Cowork tools.
Claude Upgrade Enables Direct Computer Control, Taking on OpenClaw in AI Agent Race
With these enhancements, Claude can now interact with a user’s device by opening files, using a web browser, running software tools, and even editing spreadsheets.
When given a task, the system first attempts to use connected applications such as Slack or Calendar, but if no integration is available, it operates directly on the device by navigating the screen—clicking, pointing, and completing actions itself.
At the same time, the company clarified that Claude always seeks user permission before taking control of the system.
To demonstrate these features, the company shared a video on YouTube highlighting how the system works in real scenarios. In one example, a user sends a request from their phone asking their computer to export a pitch deck and attach it to a meeting invitation scheduled for later in the day.
Following this, Claude activates on the laptop, completes the requested steps, and then notifies the user with details of what actions were performed.
In another scenario, a user asks whether Claude can resize 150 images stored on their desktop and add logos to each one, and the system successfully completes the task before confirming it.
Claude Expands Capabilities with Support for Repetitive and Automated Tasks
Beyond one-time actions, the tool can also manage ongoing or repetitive tasks, such as checking emails every morning when instructed.
Additionally, the company stated that Claude can retain context across multiple sessions, allowing it to handle tasks more consistently over time.
However, the developers acknowledged that the technology is still developing and may not always function flawlessly on the first attempt.
They noted, “It won’t always work perfectly: complex tasks sometimes need a second try, and working through your screen is slower than using a direct integration. We’re sharing it early because we want to learn where it works and where it falls short,” highlighting both its current limitations and the intention to improve it.
Furthermore, the company emphasized that safety measures have been implemented to reduce risks, including protections against “prompt injection,” which involves malicious inputs aimed at exploiting the AI system.
At present, this new functionality is available only to Claude Pro and Claude Max subscribers, specifically on macOS, and users must manually enable it within the desktop application settings.
This development comes shortly after the widespread popularity of OpenClaw earlier in the year, which demonstrated the strong potential of AI agents to perform tasks on behalf of users and is expected to inspire increased competition among leading AI companies.
Image Source
Comments are closed.