Anthropic PBC is finally having its own ChatGPT moment. A powerful new version of its Claude chatbot can now take actions on a computer, and the broad repercussions of that advance are impossible to predict. But they may well be painful for some white-collar workers.
The San Francisco-based startup, which has raised more than $30 billion since its founding in 2021, frames itself as the most safety-conscious of the artificial-intelligence developers. Its founders split from OpenAI after deeming its direction too commercial, and that cautious approach has helped the company cultivate a promising business selling trustworthy AI chatbots to companies.
About a year ago, Anthropic released Claude Code for software developers. It quickly became one of the best-regarded AI coding tools among engineering teams at companies like Netflix Inc., Spotify and Nvidia Corp., and was on pace to achieve $1 billion in annual revenue six months after launch, according to Anthropic. Programmers liked that Claude Code could plug directly into a computer’s terminal, the text-based interface where they type commands, rather than click through menus. That meant Claude Code could see their files, interrogate their source code and execute tasks autonomously, like a colleague sitting at their machine rather than a tool they copied-and-pasted from.
Boris Cherny, who runs Claude Code at Anthropic, tells me that despite being a prolific programmer, he hasn’t written a single line of computer code in the last two months. Like his team members, he’s using plain-English prompts with AI. “That’s what coding is nowadays,” he says. Two other technology startups I spoke to concurred. “Most of our engineers are not writing code anymore,” says Lawrence Jones, founding engineer at London-based Incident.io, which troubleshoots software problems and was an early adopter of Claude Code. “They’re talking through solutions with Claude.”
When Anthropic’s engineers noticed that people without programming experience were using the tool, they were struck with an idea: Perhaps Claude Code could also excel at using a computer. Computer code is a near-perfect mirror of sights, sounds and actions in the digital world, so why couldn’t an AI programmer be a computer operator too?
Over ten days, Anthropic’s engineers used Claude Code to build a more user-friendly version of itself, one that had the look and feel of a regular chatbot. They didn’t write lines of code but rather used every day language to guide Claude Code in the black terminal of their laptops. The result was Claude Cowork, which was rolled out last week and is now available to anyone with a $20-a-month subscription to Claude, as well as the company’s enterprise customers. Anthropic has emphasized it’s still a “research preview” and may have rough edges.
But Cowork has put Anthropic at an inflection point, where the hype around its coding tool could create momentum for a more mainstream breakout. Claude Code has already gone viral in tech circles, becoming a hot topic on X, Discord and at tech confabs. It’s earned the praise of non-coders too. Ben Guerin, who runs a media agency in London, used it to spin up a website for finding pubs that were struggling financially. He described what he wanted in plain language and Claude Code built the entire site, working in the background on the project during his morning commute and while he was at his day job, with minimal babysitting. The site gained about 400,000 visits within about two weeks.
"Claude Code can access stuff that Claude can't," Guerin says. And so too can the new Cowork. I recently pointed the app to a folder of interviews on my laptop, which I used to write my last book. Within minutes it had created a Power Point presentation of key trends from my files and a spreadsheet of my contacts and their areas of expertise. It also answered some direct messages on LinkedIn for me. Chatbots like ChatGPT or Perplexity can only work with what you paste into them, but Cowork can navigate your computer files, including documents, folders and applications, as well as on some online platforms if you install the relevant browser extension.
Technology leaders like Salesforce Inc. Chief Executive Officer Marc Benioff spent much of last year building hype around AI agents, bots that not only generated information but could also take action — like a customer-service chatbot making a restaurant booking. So far they’ve been a letdown thanks to reliability issues and high costs.
But Claude Cowork appears to be one of the first general-purpose AI agents, and arguably a working example of what Microsoft Corp. could have built with its own AI tool Copilot, which can operate across Microsoft Office. The company has been working on Copilot for years with an army of developers, but Cowork, built in a week-and-a-half by a small team using AI, seems broadly more capable. In many ways it is simply a more user-friendly version of Claude Code, but like its predecessor, has been getting rave reviews from tech influencers and on social media.
People have used it to organize files on their desktop, go through their Gmail inboxes to unsubscribe from marketing emails and cross reference information between different websites. The growing momentum feels like the early, viral spread of ChatGPT as the public experimented with a tool that could replace a researcher. In this case, Claude Cowork could displace many more types of employees. That raises difficult questions about how safety-conscious Anthropic could end up exacerbating the job destruction that OpenAI seems to have triggered with ChatGPT.
In some ways, narrowing the skills gap between software developers and regular folks puts more opportunity in the hands of people like Guerin, who built his pub website in a matter of hours. But it strikes a worrying note for coders, and people in other fields too. At a recent launch of Anthropic’s new tool for doctors — which can do paperwork and help make a diagnosis — CEO Dario Amodei said coding was unique in being able to “generalize to other domains,” like healthcare. As Cowork gains momentum, we may see Amodei forge into finance, media or other fields. But it seems even the people at Anthropic don’t know what the impact of that will be.
“It’s been very surprising,” says Cherny of Claude Code’s success, adding that his team observes their users to make adjustments on the fly, alongside their own ideas for direction. For all Anthropic’s good intentions, the biggest surprise may be that its ultra-safe AI tools may be among the most disruptive to the employment market.
A message from Advisor Perspectives and VettaFi: Discover something new! Click here to register for our upcoming webcasts.
Bloomberg News provided this article. For more articles like this please visit
bloomberg.com.
Read more articles by Parmy Olson
