How Claude Code’s Creator Boris Cherny Actually Uses It

A practical look at how Boris Cherny, Claude Code’s creator, uses loops, parallel agents, CLAUDE.md, automated reviews, and phone-based coding

Boris Cherny created Claude Code, the system now changing how developers explore repositories, write features, fix bugs, review pull requests, run tests, and manage entire coding workflows. Then he pushed it further. He stopped writing code by hand. He stopped opening his IDE. Eventually, he even moved beyond prompting Claude one task at a time.

Now he builds loops.

Agents monitor his pull requests, repair failing CI, review code, gather feedback, run experiments, and return with evidence. Some work from separate branches. Some challenge the work of other agents. Some keep running while he is away from his computer. Boris increasingly supervises all of this from his phone.

This is not another developer showing a clever demo.

This is the creator of Claude Code revealing what coding looks like after the tool becomes powerful enough to change your entire working method.

And that is exactly why I went deep into his material.

I watched eight of Boris Cherny’s latest workshops, live coding sessions, technical interviews, and company case studies:

The Future of Work with Claude
Fireside Chat with Boris Chern
DoorDash Gave Every Employee Claude Code
How Spotify Runs Agents Across 20M+ Lines of Code
Live Coding with Boris Cherny and Jarred Sumner
Practical Tips on How to Use Claude Code
Why Coding Is Solved and What Comes Next
Claude Code and the Future of Engineering

These are not recycled “10 Claude Code tips” videos. They are the closest thing we currently have to watching the people building this new way of working explain it while it is still being invented.

The conversations move from Boris’s own daily setup to Spotify running agents across more than 20 million lines of code, Bun using agents to reproduce bugs and open tested pull requests, DoorDash giving Claude tools across the company, and Anthropic moving from individual prompts toward loops that continue working on their own.

I also worked through Claude’s latest official documentation, the new guides on loops, goals, parallel agents, skills, dynamic workflows, model selection, effort settings, automated review, and long-running cloud routines.

Altogether, this became close to nine hours of videos, transcripts, documentation, examples, and command checking.

I went through that material because this is where our practical guides should come from not feature announcements copied into a list, but the people actually building and using these systems at the highest level.

What follows is the part worth saving.

In the next 18 minutes, you will see how Claude Code’s own creator actually uses it: how he starts inside an unfamiliar repository, what he puts in CLAUDE.md, how he verifies changes, how he runs parallel agents, how he turns repeated prompts into loops, how automated PR workflows work, and how all of these pieces fit together into one practical setup.

This is not a tour of Claude Code.

It is the working method behind it.