As software WIRED reporter, I’ve tested a lot of shitty agents over the last few years. These experiences reveal a consistent pattern of generative AI startups overpromising and underdelivering when it comes to “agent” helpers — programs designed to take control of your computer, chores, and digital affairs to free up time for more critical things. But the bots I installed on my laptop had difficulty performing even basic tasks. They just didn’t work.
This destitute track record makes Anthropic’s newest agent, Claude Coworknice surprise. When I tested it by running some of the basic and intermediate demos the company suggested 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, run reports, and even hijack your browser to search the web or organize your Gmail inbox. When it comes to file management and desktop interfaces, this tool seems to be the beginning of a nice evolution in terms of user experience.
Last year, Anthropic gained a cult following for Claude Code among developers who loved its ability to understand code bases and run commands, and tech workers across San Francisco seemingly used it in their jobs all the time. But most people aren’t on the technical staff of some busy startup.
“We tried a few different ideas to see what form factor would make sense for a less tech-savvy audience who doesn’t want to use a terminal,” says Boris Cherny, head of Claude Code at Anthropic. For the past two months, Cherny has been writing all of his code using artificial intelligence. There was cooperation built using AI tools.
Released by Anthropic earlier this week as a research preview, Cowork leverages the capabilities available in the company’s coding-centric tool and makes the user experience more accessible. This tool is aimed at a broader group of non-technical users who may want to experiment with a modern way of controlling their computers but are intimidated by the command line.
First steps
Reece Rogers
Right now, Cowork is only available as a preview release for subscribers to Anthropic’s $100/month plan, which is a common release strategy for generative AI companies to soft-launch modern features to early adopters.
Felix Rieseberg, a technical fellow at Anthropic who focuses on Cowork, says he uses it for expense reports and file conversions. “If this PDF is too large, make it smaller,” he advises. “Turn these 20 JPEG files into one PDF file. Give me a report on all this stuff.” Rieseberg is excited about how to do it more advanced users is already experimenting with sophisticated applications, but considers the simplest file-centric applications to be his “favorite” uses for research preview.
This early release is version confined Claude on Macwith the possibility of wider implementation in the future. And while you can apply it to interact with files on your computer, Cowork requires an internet connection to work. The Cowork tab appears next to the Chat and Code tabs in the Claude app for macOS. User sessions are marked as “tasks” rather than “chats”.
What about security threats?
The biggest reason not to try Cowork is the ongoing security risk inherent in these types of agents. Like most agents, Cowork is vulnerable to injection attacks – secret messages hidden across the Internet that try to trick AI tools and distract them from their tasks. You should not expose sensitive data to a tool that may be compromised in this way.
“Because Claude can read, write, and permanently delete these files, you should exercise caution when granting access to sensitive information such as financial documents, credentials, or personal information,” Anthropic says online support page. He suggests backing up your most critical files and creating a dedicated folder filled with non-sensitive information that you want Claude to have access to.