Hands On With Anthropic’s Claude Cowork, an AI Agent That Actually Works

as a software WIRED reporter, I’ve tested a lot of lousy agents over the years. These experiences highlight a consistent pattern of generic AI startups overpromising and underdelivering when it comes to these “agent” assistants — programs designed to take over work and digital work by taking control of your computer, freeing up your time for more important things. But the bots I installed on my laptop would have difficulty completing even basic tasks. He didn’t even work.

This poor track record makes Anthropic’s newest agent, Claude Cowork, a pleasant surprise. When I tested it by running it through some basic and intermediate demos suggested by the company in addition to my own commands, it worked quite well – especially for software that’s still in beta. It can do things like organize files into folders, convert file types, generate reports, and even search the web or take over the browser to organize the Gmail inbox. When it comes to file management and computer interface, this tool seems like the beginning of a pleasant user experience evolution.

Last year, Anthropic nurtured a cult following for its cloud code tool among developers who loved its ability to understand the codebase and run commands, with tech workers across San Francisco using it for their work all the time. But most people aren’t technical staff members of some enthusiastic startup.

“We tried a bunch of different ideas to see what form factor would make sense for a less technical audience who doesn’t want to use a terminal,” says Boris Cherny, Anthropic’s head of cloud code. For the past two months, Cherney has written all of his code with AI. Cowork was built using AI tools.

Released as a research preview earlier this week by Anthropic, Cowork takes the capabilities available in the company’s coding-focused tools and makes the user experience more accessible. This tool is designed for a broad group of non-technical users who want to experiment with new ways of controlling their computers but are intimidated by the command line.

launch

Image can contain pages and text

reece rogers

Right now, Cowork is only available as part of a research preview for customers on Anthropic’s $100 per month plan, a common release strategy for generic AI companies soft-launching new features for early adopters.

Felix Rieseberg, a member of Anthropic’s technical staff who focuses on Cowork, says he uses it to enter expense reports and perform file conversions. “If this PDF is too big, make it smaller,” he says. “Convert these 20 JPGs to a PDF. Make a report for me about all these things.” Rieseberg is excited by how more advanced users are already experimenting with complex applications, but sees the simplest, file-centric applications as “my favorite” use of Research Preview.

This initial release is limited to the cloud on Mac, with a wider rollout possible in the future. And even though you can use it to interact with files on your computer, Cowork requires an Internet connection to run. The Cowork tab appears next to the “Chat” and “Code” tabs in the Cloud app for macOS. User sessions are labeled as “Tasks” rather than “Chat”.

What about security risks?

The biggest reason to try Cowork is the security risks inherent in these types of agents. Like most agents, Cowork is vulnerable to prompt injection attacks, secret messages hidden online that attempt to trick AI tools and distract them from tasks. You should not expose sensitive data to tools that could be compromised in this way.

Anthropic’s online help page reads, “Since the cloud can read, write, and permanently delete these files, be cautious about providing access to sensitive information such as financial documents, credentials, or personal records.” It suggests creating a dedicated folder filled with non-sensitive information that you want the cloud to be able to access and save backups of important files.



<a href

Leave a Comment