Exploring Anthropic’s Claude Cowork: An AI Assistant That Delivers Results

As a software journalist at WIRED, I’ve evaluated numerous ineffective agents over the last few years. These experiences reveal a recurring trend among generative AI startups: they often make grand promises but fail to deliver when it comes to these “agentic” assistants—programs intended to take charge of your computer, handling tasks and digital errands to free up your time for more significant matters. However, the bots I installed on my laptop struggled to perform even simple functions. They simply didn’t work.

This disappointing history makes Anthropic’s newest agent, Claude Cowork, a refreshing surprise. In my testing, using both the basic and intermediate demos recommended by the company along with my own commands, it performed quite well—especially considering it’s still in beta. It can organize files into folders, convert file types, generate reports, and even take control of your browser for searches or to tidy up a Gmail inbox. In terms of file management and computer interfaces, this tool seems to signal the beginning of a positive evolution in user experience.

Last year, Anthropic garnered a devoted following for its Claude Code tool among developers who appreciated its proficiency in understanding codebases and executing commands, with tech employees throughout San Francisco frequently utilizing it for their tasks. Yet, most users aren’t part of some trendy startup’s technical team.

“We explored various concepts to determine what form would be suitable for a less technical audience that doesn’t wish to engage with a terminal,” states Boris Cherny, Anthropic’s head of Claude Code. For the past two months, Cherny has written all of his code utilizing AI. Cowork was developed using AI tools.

Released by Anthropic earlier this week as a research preview, Cowork adapts the capabilities found in the company’s coding-focused tool, making the user experience more accessible. This tool is tailored for a broader audience of non-technical users, who might be interested in experimenting with a new method of controlling their computers but are intimidated by a command line.

Getting Started

Reece Rogers

Currently, Cowork is only accessible as part of a research preview for subscribers of Anthropic’s $100-a-month plan, a common strategy among generative AI firms launching new features to early adopters.

Felix Rieseberg, a technical staff member at Anthropic focused on Cowork, mentions he uses it for tasks like filing expense reports and converting files. “If this PDF is too large, make it smaller,” he states. “Combine these 20 JPGs into one PDF. Create a report about all these items.” Rieseberg is enthusiastic about how more advanced users are already testing out complex applications, but he considers the straightforward, file-centric uses as “my favorite” applications of the research preview.

This initial release is limited to Claude on Mac, with a broader rollout likely in the future. While you can use it to manage files on your computer, an internet connection is necessary for Cowork to operate. The Cowork tab is found next to the “Chat” and “Code” tabs within the Claude app for macOS, and user sessions are referred to as “tasks” instead of “chats.”

What About the Security Risks?

One major concern regarding the use of Cowork is the ongoing security risks associated with these types of agents. Like many agents, Cowork is vulnerable to prompt injection attacks, where secret messages can manipulate AI tools and divert them from their intended tasks. Sensitive data should not be shared with a tool that could be compromised in this manner.

“Since Claude can read, write, and permanently delete these files, exercise caution when granting access to sensitive information such as financial records, credentials, or personal documents,” warns Anthropic on their online support page. They recommend creating a dedicated folder filled with non-sensitive information for Claude to access and ensuring backups of important files are stored safely.