Software As a reporter for WIRED I have tested many shitty agents These experiences reveal a consistent pattern of behavior over the past two years. The experiences reveal a pattern that is consistent. generative AI These startups are notorious for overpromising but underdelivering. “agentic” helpers—programs designed to take control of your computer, performing chores and digital errands to free up your time for more important things. The bots on my laptop are not malicious. would struggle Even basic tasks are difficult to accomplish. The system just did not work.
Anthropic is the latest agent to have a bad track record. Claude CoworkA pleasant surprise. When I tested it by running it through some basic and intermediate demos the company suggested in addition to my own commands, it worked fairly well—especially for software It’s in beta. The tool can organize files in folders, create reports and convert file formats. It will even search the internet or clean up your Gmail Inbox. It feels like a step forward in the evolution of computer and file interfaces.
Anthropic cultivated a following of cult followers for its Claude Code It is a favorite among San Francisco’s tech workers, who use it to do their jobs almost constantly. The majority of people, however, are not members of a startup’s tech staff.
“We tried a bunch of different ideas to see what form factor would make sense for a less technical audience that doesn’t want to use a terminal,” Boris Cherny is Anthropic’s Head of Claude Code. Cherny wrote all his code using AI for the last two months. Work was built using AI tools.
Released by Anthropic Cowork, which was unveiled earlier in the week as part of a research preview by the company, makes the tools available to the tool’s users more accessible. This tool has been designed for the non-technical user who wants to explore a new method of control their computer, but is intimidated by it. command line.
Getting Started
Reece Rogers
Cowork, at the moment, is only available as a preview of research to Anthropic subscribers on their $100/month subscription plan. common release strategy Early adopters of generative AI can be softly introduced to new features by companies.
Felix Rieseberg is a technical employee at Anthropic, who works with Cowork. He uses the software to convert files and file expenses. “If this PDF is too big, make it smaller,” He says. “Turn these 20 JPEGs into one PDF. Make me a report about all of these things.” Rieseberg was excited to see how more advanced users The most simple, straightforward applications that focus on files are his favorites. “favorite” The research preview can be used in many ways.
Early release only applies to certain products Claude on MacCowork is expected to be rolled out more widely in future. Cowork requires an internet connection to work, even though it can be used to interact with documents on your computer. In the Claude App, the Cowork button is located next to Chat and Code. macOS. The user sessions are labelled as “tasks” Even though “chats.”
What about the security risks?
Cowork’s ongoing costs is the main reason why people don’t use it. security The agents themselves are at risk. Cowork, like most agents, is vulnerable to prompt injection, which are secret messages that have been hidden on the internet to try and trick AI tools into deviating from their tasks. Do not expose sensitive data to an AI tool that is susceptible.
“Since Claude can read, write, and permanently delete these files, be cautious about granting access to sensitive information like financial documents, credentials, or personal records,” reads Anthropic’s online support page. You can save backups for important files, and create a special folder that contains non-sensitive information.
