Claude Cowork: The $100 Beta Tool That Sorts Your Desktop But Can’t Fix Your Life

Anthropic’s Claude Cowork proves AI agents might finally work—but only if you’re willing to tolerate beta software and pay $100/month. The macOS-only tool offers file organization, browser automation, and partial task completion, but its limitations are glaring. Users must grant explicit permissions for folder access and browser actions, and the system’s reliance on internet connectivity creates bottlenecks for offline workflows.

When I tested it by running it through some basic and intermediate demos... it worked fairly well—especially for software that’s still in beta. The AI can sort screenshots by month or generate reports from PDFs, but it stumbles on critical tasks like purchasing movie tickets.

"If this PDF is too big, make it smaller... Make me a report about all of these things," the system might request—only to fail when asked to finalize a purchase. "We use a virtual machine under the hood... Claude literally cannot see that folder," a developer explained, highlighting the security sandbox that also hinders functionality.

Security Risks and Limitations

The biggest reason for not trying out Cowork is the ongoing security risk inherent in these kinds of agents. Prompt injection attacks and potential malware exposure via browser interactions remain unresolved.

"The bot messing up is still a real possibility as Anthropic works through bugs," one tester noted after the AI misfiled documents and crashed during email archiving. While vendors tout a "pleasant user experience evolution," real-world testing reveals frequent manual interventions—like manually purchasing tickets after the AI finds them.