The generative AI vendor has given its cloud large language model the ability to take control of your computer and perform certain tasks on request, as it rivals the personal agent framework OpenClause.
update, Revealed in a blog dated March 23 POST enables cloud code and cloud peer Tools for interacting with the user’s machine, opening files, working with browsers, running tools, and filling out spreadsheets.
Once prompted, the cloud will first look for connected apps like Slack, Calendar or other integrations, but if there’s no connector for the required tools, it’ll work locally on the device, pointing, clicking and navigating to whatever is on the screen to perform the requested task – although according to the vendor, it always asks for permission first.
A video released by the seller, posted on youtubeDemonstrated the new possibilities offered by the cloud.
One of the examples shown here is a user sending a request to export a pitch deck from their phone to their computer and then attaching it to an invitation to a scheduled meeting later that day.
The cloud activates on the laptop, performs the task as requested, then returns a message to the user to inform them of what specifically has been done.
The second use case involved sending a user a message asking if the cloud could resize 150 photos on their desktop and add a logo to each. Again, Cloud is shown doing as he was told and then sending confirmation.
The agent is also capable of handling “repetitive” tasks if requested, such as scanning email every morning, with Anthropic saying it is able to remember context across different sessions.
However, Anthropic acknowledged that the technology is still in its early stages and things may not get it right the first time. “It won’t always work perfectly: Complex tasks sometimes require a second attempt, and working through your screen is slower than using direct integration. We’re sharing this early because we want to learn where it works and where it falls short,” the blog says.
Anthropic also said it has created security measures to reduce the risk, for example, protection against “quick injections” – malicious inputs designed to compromise the AI.
The updated functionality is now available for Cloud Pro and Cloud Max subscribers, but only for macOS users, and must be enabled in the desktop app settings.
Cloud’s update comes after viral success open paw Earlier this year, that showed the huge potential for AI agents to complete tasks on demand from users, and is likely to pave the way for a wave of competitors from major AI companies.
