AI Is Moving Beyond Chatbots. Claude Cowork Shows What Comes Next

The DNA file had been gathering dust in Pietro Schirano’s computer for years. Then, earlier this month, he gave it to Claude Code—an “agentic coding tool” developed by Anthropic—for analysis. “I’m attaching my raw DNA file from Ancestry DNA,” he told the tool.

The AI spawned copies of itself on Schirano’s computer, each one simulating an expert in a different part of the genome—one expert on cardiovascular disease, another on aging, a third on autoimmune disease. “There were a lot of things that resonated with my life,” says Schirano, who was an engineer at Anthropic prior to founding MagicPath, an AI product design startup. “I always thought that I could deal with caffeine better than all of my friends. It was always this inside joke: I can just drink seven espressos because I’m Italian.” Claude Code’s analysis revealed that Schirano does, in fact, have a gene that allows him to metabolize caffeine better than the average person, that he’s predisposed to Alzheimer’s, and suggested supplements to take based on his genetic profile.

[time-brightcove not-tgx=”true”]

Claude Code, released in February 2025, was Anthropic’s first successful attempt at building an AI agent—a system that takes actions on the user’s behalf, rather than merely conversing in a chat interface. Claude Code can access files and programs on a user’s computer, and even run “sub-agents” for specific tasks, such as those that analyzed different parts of Schirano’s genome. It has steadily accrued a devoted following of tinkerers using it to file their taxes, design knitting patterns, and even autonomously grow a tomato plant.

Yet most people have never heard of Claude Code. That’s because the primary way of accessing the tool is through a command line interface—the old-school computer terminal that went out of fashion among the general public some time in the last millennium. That obscurity might be about to change. On Monday, Anthropic announced Claude Cowork, which the company calls “Claude Code for the rest of your work.” 

“It’s gonna blow a lot of people’s minds who are not coders,” says Martin DeVido, the developer behind the experiment using Claude Code to grow a tomato plant.

Claude Cowork aims to bring Claude Code’s agentic capabilities to a broader audience by supplying it with a friendlier user interface and hiding some of the complexity that has made Claude Code daunting to the uninitiated. The tool, initially available as a research preview for customers paying $200 a month for the Max plan, has “rough edges,” according to Felix Reiseberg, its lead engineer. One user found that the app gave her “scary error messages” and wouldn’t connect to her calendar. 

This shouldn’t be surprising: the app was built in under two weeks, mostly by Claude Code itself. It’s a sign of the times that AI tools are building themselves, and that they sometimes break as a result. However, Claude Code started life as a demo; six months later it had reached $1B in annualized revenue—reportedly around 15% of Anthropic’s total at the time.

Claude Code and Cowork are among many attempts to turn AI chatbots into agents. Other agentic coding tools, such as Cursor and Codex from OpenAI, have also found success among programmers; AI browsers including ChatGPT Atlas and Perplexity’s Comet browser seek to give AI models the ability to take actions for the user on the internet. “If Claude Code didn’t exist, I would have expected a similar [level of popularity] with Codex or Cursor,” says Jean-Stanislas Denain, a researcher at the AI research institute Epoch AI—it’s representative of other tools that give AI models agentic capabilities.

It remains to be seen whether Opus 4.5, the model powering Claude Cowork, is as good at general knowledge work as it has been for programming through Claude Code. “Opus 4.5 is certainly huge for coding,” Mantas Mazeika, research scientist at the Center for AI Safety, told TIME. However, it was only able to complete nine out of 240 human freelancer projects in the Center for AI Safety’s Remote Labor Index, which include making architectural plans and developing video games. “The main bottlenecks are specific cognitive limitations that Claude Code doesn’t address.”

“We’re not going to see from this initial deployment whether a wide variety of people who are not experienced with coding would, in fact, get value out of it,” says Denain, citing the rollout to a small handful of customers.

Still, for some, Claude Code, Cowork, and the like represent the shift from AI chatbots to agents that has been predicted since before the official release of ChatGPT. “Claude Code is just one more data point on the broader trend that’s been ongoing for years and will continue for years,” says Daniel Kokotajlo, executive director of the AI Futures Project. Kokotajlo forecast in 2021 that the “age of the AI assistant” would arrive in 2026. (His team recently predicted that artificial superintelligence—AI substantially smarter than any human—would arrive in 2034.)

“I actually think that Claude Cowork is going to be a much, much larger disruption of the economic index than anything else that we’ve seen so far, because it’s really going to impact the white-collar job,” says Schirano.