Cowork is a new feature for the AI chatbot Claude that taps into a user’s computer folder to either modify it or create new projects with the data, Anthropic said in a blog post on Monday. Cowork is currently only available to MacOS users with a $100/month Claude Max subscription.
Cowork works by having users give Claude access to a computer folder. Once in, users can then chat with Claude and have it organize all the files, rename items to match a specific naming convention or read the files inside to create spreadsheets or other documents. For example, if a user has a folder with a bunch of receipts, Cowork can organize all the files and create a spreadsheet listing all the expenses.
Don’t miss any of our unbiased tech content and lab-based reviews. Add as a preferred Google source.
Anthropic also gave Cowork the ability to link to connectors, which allows Claude to connect to other apps to create documents, presentations or other projects. Additionally, Cowork works with the Claude in Chrome extension, allowing it to complete tasks that require browser access.
“The inspiration for Cowork was watching what people did with Claude Code. We built it for coding, but people started using it for everything — taxes, managing receipts, organizing files, random life admin,” said Felix Rieseberg, member of the technical staff at Anthropic, in a statement to . “Both developers and nondevelopers started using the agent loop to get some help from Claude on their computer.”
Anthropic stresses that Claude can’t edit anything to which it hasn’t been given access. But since Cowork is still in research preview, which is essentially a beta test period, there’s a chance the AI might cause unintended consequences. The company recommends using Cowork only on nonsensitive items.
Anthropic also warns against potential prompt-injection attacks, where AIs inadvertently suck up commands from websites across the internet and perform unintended actions. The company says it’s built defenses against these possibilities. However, if an attacker uses a prompt-injection attack to seize control of Claude, which also has access to a folder on your computer containing sensitive information via Cowork, it could pose a significant problem.
Cowork by Anthropic comes as more AI models are gaining agentic capabilities. This is when an AI can do more than just chat with a user. Instead, these AIs can go out onto the internet and perform tasks on behalf of the user, such as booking travel, compiling a complex report or ordering a pizza. Tools like ChatGPT Agent and Gemini Agent are already fulfilling tasks for users.
Companies are taking AI one step further, as with Google’s Gemini 2.5 Computer Use model. This enables AI to access a computer directly, allowing it to interact with and accomplish tasks. Agentic AI is also infiltrating the retail space. Last week, Google and its retail partners announced a new open standard for online shopping that enables AIs to work across multiple retailers and payment systems.
