Anthropic has just released Cowork, which it is billing as Claude Code for non-coding work, and it turns out that it was essentially developed by Claude Code.
Boris Cherny, the creator of Claude Code at Anthropic, revealed that the newly launched Cowork product was written “pretty much all” by Claude Code itself, marking another milestone in AI’s role in building AI tools.
When asked on X about the division of labor between Claude and human developers for Cowork, Cherny’s response was unequivocal: “It was pretty much all Claude Code.”
What is Cowork?
Cowork represents Anthropic’s first major expansion of Claude Code beyond its original coding focus. The product emerged from observing how users were already stretching Claude Code’s capabilities into unexpected territories—from vacation research and building slide decks to cleaning up email, canceling subscriptions, recovering wedding photos from hard drives, monitoring plant growth, and even controlling ovens.
“These use cases are diverse and surprising,” Cherny wrote in announcing the launch. “The reason is that the underlying Claude Agent is the best agent, and Opus 4.5 is the best model.”
Cowork brings several novel features designed specifically for non-coding work, including a built-in virtual machine for isolation, out-of-the-box browser automation support, integration with all claude.ai data connectors, and the ability to ask users for clarification when uncertain. The product is currently available as a research preview for Claude Max subscribers in the macOS app.
An Accelerating Trend
This revelation continues a striking pattern in Claude Code’s development. In May 2025, Cherny disclosed that approximately 80% of Claude Code’s own codebase was being written by Claude Code itself—already a remarkable milestone. By December 2025, Cherny reported that he hadn’t opened an integrated development environment even once during an entire month, with Opus 4.5 writing around 200 pull requests comprising every single line of code he shipped.
Now, just weeks later, the announcement that Cowork was developed almost entirely by Claude Code suggests this trend has not only continued but potentially accelerated.
The Human Role
While Claude Code handled the bulk of code generation for Cowork, Cherny has consistently emphasized that human oversight remains crucial. In previous interviews, he described a workflow where Claude typically takes the first pass at coding tasks, with humans diving in when the results aren’t satisfactory or when tasks require nuanced judgment.
“I think some of the stuff has to be handwritten, and some of the code can be written by Claude,” Cherny explained on the Latent Space podcast earlier this year. “And there’s sort of a wisdom in knowing which one to pick and what percent for each kind of task.”
For intricate data model refactoring or tasks where he has strong architectural opinions, Cherny noted he still prefers to work manually rather than explaining his vision to Claude. But these cases appear to be increasingly the exception rather than the rule.
A New Development Paradigm
The development of Cowork by Claude Code represents more than just an efficiency gain—it signals a fundamental shift in software development. Engineers are transitioning from writing code to becoming architects, reviewers, and orchestrators of AI-generated code.
Cherny has noted that this mental shift represents one of the biggest challenges, even for practitioners at the forefront of AI development. “It takes significant mental work to re-adjust to what the model can do every month or two, as models continue to become better and better at coding and engineering,” he said.
Interestingly, he’s observed that newer engineers and recent graduates may actually have an advantage in this environment. Without legacy assumptions about AI limitations formed during the era of less capable systems, they’re able to leverage current technology more effectively.
Still Just Getting Started
Despite the dramatic nature of these changes, Cherny maintains the industry is only at the beginning of this transformation. “Software engineering is changing, and we are entering a new period in coding history,” he wrote in December. “And we’re still just getting started.”
The fact that Cowork—a product designed to expand AI capabilities into new domains—was itself built almost entirely by AI offers a compelling demonstration of both the technology’s current capabilities and its potential trajectory. As Cherny’s experience shows, the line between AI as tool and AI as co-creator continues to blur, raising profound questions about the future of software development and knowledge work more broadly.
For now, Cowork stands as both a new product offering and a proof of concept: evidence that AI agents are already capable of building the next generation of AI agents, with humans increasingly playing the role of guides and reviewers rather than primary creators.