Boris Cherny is the creator and head of Claude Code, a tool that started as an internal prototype at Anthropic and, by his account, within a year came to account for roughly four percent of all public commits on GitHub. What struck me most in his interview, and what do I think follows from it?
What the creator of Claude Code says about how software is built today
In April 2026, Boris Cherny described how he works with thousands of autonomous AI agents.
1. The model writes the code, not the person
Cherny says he no longer writes a single line of code by hand. The model generates all of it for him. He considers programming a largely solved problem. This is not a claim about the future but a description of how he works right now.
More interesting than the statement itself is what follows from it: if the model writes the code, what people do changes. Cherny describes his role as assigning work, reviewing it, and deciding what to work on. Not writing.
2. He works from his phone, in parallel
He describes most of his work as assigning tasks from his phone, with several sessions running at the same time. The model works in the background, and he comes back to the results when they are done.
The key element, he says, is automated loops and routines: scheduled jobs that start on their own and take care of things a person would otherwise have to keep in mind, such as maintaining test infrastructure, continuously merging changes into work in progress, and collecting feedback. Nobody starts them. They just run.
3. Inside Anthropic, there is no hand-written code
According to Cherny, models at Anthropic generate everything, including database queries. Agents talk to each other over the company Slack when they need to clarify something in a task. They ask other agents, not people.
He does not see the company’s edge primarily in the technology, which others have equal access to. He sees it in the fact that the company has changed its organizational structure and processes so that this way of working can actually run. In my view, this is the most important point of the whole interview: the competitive advantage is not in the model, but in how the company is built around it.
4. Everyone will write code
Cherny expects development teams to shift from specialists to generalists. Code will be written by product managers, designers, researchers, everyone who today hands work to developers.
He compares the ability to create software to literacy, and its spread to the invention of the printing press. In his view, the best software for a given field will not be written by engineers but by practitioners, such as accountants, lawyers, and doctors, because the coding itself will be the easiest part and what really matters is knowing the field.
5. The market will shift in favor of those starting from scratch
According to Cherny, competitive advantages built on established processes or high switching costs are weakening. He expects a significant increase in startups that can build a company from the ground up with AI and compete with corporations weighed down by legacy systems and old ways of working.
6. What comes next
For automation in office tools, such as documents, CRM, and email, he sees the MCP protocol as key, because it lets models work with systems regardless of where they run. He sees further growth in models’ ability to operate a computer directly and in massive parallelization of tasks.
What I take away from it
Cherny describes an environment where thousands of agents are business as usual. Most companies I talk to are far from that, and I don’t mean that as criticism. Getting AI into an organization so that it actually works is harder than interviews like this make it look.
But one thing is clear to me. Anthropic does not have thousands of agents because it has better technology. It has them because it built the company so that it is possible to work with them: so that it is clear what each one does, who is responsible for it, and how they coordinate with each other. Without that, thousands of agents would not be an advantage. They would be chaos.
In my view, this is the part of the interview that applies to every company, not just those that develop software. Technology can be bought. The ability to manage it cannot.
