Credit: Anthropic

Anthropic’s agentic tool Claude Code has been an enormous hit with some software developers and hobbyists, and now the company is bringing that modality to more general office work with a new feature called Cowork.

Built on the same foundations as Claude Code and baked into the macOS Claude desktop app, Cowork allows users to give Claude access to a specific folder on their computer and then give plain language instructions for tasks.

Anthropic gave examples like filling out an expense report from a folder full of receipt photos, writing reports based on a big stack of digital notes, or reorganizing a folder (or cleaning up your desktop) based on a prompt.

A lot of this was already possible with Claude Code, but it might not have been clear to all users that it could be used that way, and Claude Code required more technical know-how to set up. Anthropic’s goal with Cowork is to make it something any knowledge worker—from developers to marketers—could get rolling with right away. Anthropic says it started working on Cowork partly because people were already using Claude Code for general knowledge work tasks anyway.

I’ve already been doing things similar to this with the Claude desktop app via Model Context Protocol (MCP), prompting it to perform tasks like creating notes directly in my Obsidian vault based on files I showed it, but this is clearly a cleaner way to do some of that—and there are Claude Code-like usability perks here, like the ability to make new requests or amendments to the assignment with a new message before the initial task is complete.

That said, there are some causes for concern. Generally speaking, you expect a Claude Code-using developer or a hobbyist geeky enough to play around with MCP to understand the risks in what they’re doing. Less technical users might not have that foresight.

Anthropic’s announcement about Cowork names a few potential concerns up front. First, you have to be careful with how you word your prompts, as a vague one (or, frankly, just incredibly bad luck) can lead the agent to do destructive things like unexpectedly delete files.

There’s also the very real, seemingly unsolvable risk of prompt injection attacks.

Given all of that, Cowork is currently available only as a research preview to Max subscribers. There’s no word about when it might see a wider release.

This isn’t the only effort Anthropic has made to branch out beyond its established footprint among coders this week. Yesterday, the company also announced Claude for Healthcare, competing with OpenAI’s similar announcement of a health management tool for ChatGPT.