When Boris Cherny, the mastermind behind Claude Code at Anthropic, shared his workflow on X, it sent shockwaves through the developer community. What started as a simple look into his terminal setup quickly morphed into a detailed manifesto on the future of software development—one that has left many in Silicon Valley scrambling to keep up.
A New Era in Coding
According to industry insiders, Cherny's insights are nothing short of revolutionary. "If you're not reading the Claude Code best practices straight from its creator, you're behind as a programmer," wrote Jeff Tang, a prominent voice in the developer community. Kyle McNease echoed this sentiment, suggesting that Cherny's updates mark a potential "ChatGPT moment" for Anthropic, a company already making waves in the AI landscape.
What's fueling this excitement? It's the paradox of Cherny's approach. Despite being remarkably simple, his workflow allows him to wield the productivity of a small engineering department. One user even commented that working with Cherny's setup felt more like a game of Starcraft than traditional coding—highlighting a fundamental shift from typing syntax to orchestrating autonomous units.
Mastering Parallelism
The most eye-opening aspect of Cherny's workflow is his method of managing multiple AI agents simultaneously. Unlike the traditional linear coding process—writing a function, testing it, and repeating—Cherny's approach resembles that of a fleet commander strategizing in real-time.
"I run 5 Claudes in parallel in my terminal," Cherny explained. "I number my tabs 1-5, and use system notifications to know when a Claude needs input."
This system allows him to juggle numerous tasks: while one Claude runs tests, another handles refactoring, and yet another drafts documentation. He even extends this approach to his browser, employing 5-10 additional Claudes on claude.ai, using a "teleport" command to switch between sessions on his local machine and the web.
Quality Over Speed
In a field that often prioritizes speed, Cherny's choice of model raises eyebrows. Rather than opting for the fastest AI, he relies on Anthropic's heaviest and slowest model, Opus 4.5. Cherny argues that this model, despite its latency, actually enhances productivity.
"It's the best coding model I've ever used, and even though it's bigger & slower than Sonnet, since you have to steer it less and it's better at tool use, it is almost always faster than using a smaller model in the end," he said.
This perspective challenges the tech community's obsession with minimizing latency. Cherny suggests that the real bottleneck in AI development is not the speed of generative processes—it's the time developers spend rectifying AI errors. By investing in a smarter model upfront, teams can save time on corrections later.
Transforming Mistakes into Lessons
Another fascinating and crucial component of Cherny's workflow is his team's approach to addressing AI forgetfulness. Large language models often lack the ability to remember specific coding styles or architectural decisions across sessions. To mitigate this, Cherny's team maintains a single file titled CLAUDE.md in their git repository.
"Anytime we see Claude do something incorrectly we add it to the CLAUDE.md, so Claude knows not to do it next time," he shared.
This method effectively turns their codebase into a self-correcting entity. When a human developer identifies an error during a code review, they don’t just fix it—they also update the AI's instructions, turning every mistake into a rule.
Automation: The Name of the Game
Cherny further streamlines his workflow through rigorous automation of repetitive tasks. By utilizing slash commands—custom shortcuts embedded in the project’s repository—he can execute complex operations with a simple keystroke. One particularly lauded command is /commit-push-pr, which automates the tedious process of version control.
In addition, he employs subagents, specialized AI personas designed for specific phases of the development lifecycle. For instance, he has a code-simplifier for architectural cleanup and a verify-app agent that conducts end-to-end tests before anything is finalized.
Verification Loops: Raising the Bar
Perhaps the most compelling aspect of Cherny's workflow is the verification loop, which has played a significant role in Claude Code reaching an impressive $1 billion in annual recurring revenue. Cherny emphasizes that Claude isn't just a text generator; it's a tester as well.
"Claude tests every single change I land to claude.ai/code using the Claude Chrome extension. It opens a browser, tests the UI, and iterates until the code works and the UX feels good," he explained.
This capability is game-changing—providing a way for the AI to verify its output through browser automation, running commands, or executing test suites. The result is a marked improvement in code quality, often yielding a 2-3x increase in effectiveness.
The Future of Software Engineering
The overwhelming reaction to Cherny's insights indicates that we may be standing on the brink of a pivotal transformation in software engineering. For years, many viewed AI coding tools as simple autocomplete features. However, Cherny has shown us that AI can serve much more substantial roles, acting as a robust operating system for labor itself.
As Jeff Tang succinctly put it, "Read this if you're already an engineer... and want more power." The tools to multiply human output by a factor of five are already in play. All it takes is a shift in perspective—seeing AI not just as an assistant but as a workforce ready to take on complex tasks.
At the end of the day, the programmers who embrace this new mindset will not only boost their productivity; they'll find themselves playing a whole new game—leaving those who cling to traditional methods still typing away.
Sam Torres
Digital ethicist and technology critic. Believes in responsible AI development.
