Boris Cherny Claude
Boris Cherny, the creator of Claude Code, revealed his full AI coding setup at Sequoia running 100+ agents, 150 PRs a day, all from his iPhone.
The engineer behind the world’s most powerful AI coding tool just pulled back the curtain on how he actually uses it, and the answer is more radical than most developers expect.
Boris Cherny is not a household name outside of software engineering circles, but inside them, he is building something that is quietly rewriting the rules of how code gets written. Cherny is the creator of Claude Code, the agentic coding tool developed by Anthropic that has rapidly become one of the most talked-about products in the AI industry. At a recent Sequoia AI session, Cherny sat down for a candid 15-minute conversation that offered the clearest window yet into his personal workflow. His philosophy on software development, and his bold predictions for the future of building technology.
The headline figure is striking: 100% of Boris Cherny’s code is written by Claude Code. Not most of it. Not the boilerplate. All of it. And he is running somewhere in the range of a few hundred agents at any given moment, with thousands more doing deeper overnight work while he sleeps.
From Accidental Invention to Industry-Defining Product
Claude Code was not born from a grand strategic plan. Boris Cherny joined Anthropic in late 2024 as part of an internal incubator called Anthropic Labs, a small innovation team that also produced the Model Context Protocol (MCP) and the Claude desktop app. The mandate was simple: build the thing you believe should exist.
At the time, the state of the art in AI-assisted coding was tab completion. A developer opened their IDE, pressed a key, and completed one line at a time. Cherny and his team believed the underlying model capability had already outgrown that approach. The question was not whether an agent could write entire programs. It was whether the product could be built in time to meet the model where it was heading.
“We were building for the next model,” Cherny explained at the Sequoia session. “We knew it wouldn’t have product-market fit for six months.”
That patience paid off. Claude Code’s exponential growth trajectory began with the release of Opus 4 in May 2025, and has continued to accelerate with every subsequent model release through 4.5, 4.6, and now 4.7.
Inside the Setup of Loops, Agents, and a Phone
Perhaps the most revelatory part of Boris Cherny’s Sequoia appearance was his description of his actual daily workflow, one that most engineers would find almost unrecognizable.
Cherny manages most of his work from his iPhone, running five to ten Claude Code sessions simultaneously, each containing multiple active agents. On a particularly productive day last week, he merged 150 pull requests in a single day. He describes that as a personal record, achieved largely as an experiment to see how far the system could be pushed.
The feature he is most enthusiastic about right now is called the loop, accessed via a /loop command. Which uses a cron-style job scheduler to keep agents running on repeat tasks indefinitely. Cherny currently maintains dozens of active loops handling tasks including:
- Monitoring and fixing continuous integration failures
- Auto-rebasing pull requests
- Scraping and clustering product feedback from social media every 30 minutes
- Babysitting open pull requests through the review cycle
Anthropic has since extended this capability with a server-side feature called Routines, which keeps loops running even after a laptop is closed or a session ends.
A Generalist Future and the End of Silos
Beyond his own setup, Boris Cherny offered a pointed take on where software teams are heading. His view is that the traditional model of narrow technical specialists is giving way to something broader and more cross-disciplinary.
On the Claude Code team itself, every member writes code. Also including the product manager, engineering manager, designers, data scientists, and user researchers. The barriers between disciplines, Cherny argues, are dissolving because the cost of writing software is collapsing.
This shift, he believes, will be most consequential at the startup level. As AI makes software development dramatically cheaper and faster, small founding teams will be able to build and compete. Especially, at a scale that previously required large organizations. The advantage will go to teams that adopt AI-native workflows from day one. Notably, rather than incumbents who must retrain existing workforces and navigate organizational resistance.
The Broader Signal
Boris Cherny’s biography reads like a quiet accumulation of technical depth, from writing a guide to BASIC programming for TI-83 calculators in middle school. To authoring a widely used textbook on TypeScript. Especially, to now overseeing the tool that many believe represents the leading edge of human-AI collaboration in software engineering.
What makes his Sequoia session worth watching is not just the product roadmap it implies. It is the lived demonstration that the future he is describing is already operational. Running right now, from an iPhone, a few hundred agents at a time.
