When the guy who built the most advanced coding agent in the world casually drops his personal workflow on X, people pay attention. For the past week, Boris Cherny, creator and head of <a href="https://data.allwinchina.org/ai-tools/claude-code/" title="Claude Code review”>Claude Code at Anthropic, has been the center of a firestorm in the engineering community.
What started as a simple thread about his terminal setup has turned into a viral manifesto on how software development is going to work from now on. Industry insiders are calling it a watershed moment for Anthropic. Jeff Tang, a well-known developer voice, put it bluntly: “If you’re not reading the Claude Code best practices straight from its creator, you’re behind as a programmer.”
Kyle McNease went even further, saying Anthropic is “on fire” and potentially facing “their ChatGPT moment.” That’s not hyperbole when you see what Cherny is actually doing.
The wild part? His workflow is surprisingly simple. Yet it lets a single human operate with the output capacity of a small engineering department. One user who tried it said the experience “feels more like StarCraft” than traditional coding. You’re not typing syntax anymore. You’re commanding units.
Running five AI agents at once turns coding into a real-time strategy game
The most striking thing about Cherny’s disclosure is that he doesn’t code linearly. The traditional “inner loop” of development goes: write a function, test it, move to the next. That’s not how he works at all.
“I run 5 Claudes in parallel in my terminal,” Cherny wrote. “I number my tabs 1-5, and use system notifications to know when a Claude needs input.”
He uses iTerm2 system notifications to manage five simultaneous work streams. While one agent runs a test suite, another refactors a legacy module, and a third drafts documentation. He also runs “5-10 Claudes on claude.ai” in his browser, with a “teleport” command to hand off sessions between the web and his local machine.
This validates the “do more with less” strategy Anthropic President Daniela Amodei talked about earlier this week. While competitors like OpenAI are building trillion-dollar infrastructure, Anthropic is showing that better orchestration of existing models can give you exponential productivity gains.
The counterintuitive case for choosing the slowest, smartest model
Here’s where Cherny goes against the grain. In an industry obsessed with speed and low latency, he exclusively uses Anthropic’s heaviest, slowest model: Opus 4.5.
“I use Opus 4.5 with thinking for everything,” Cherny explained. “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.”
This is a critical insight for anyone running engineering teams. The bottleneck in modern AI development isn’t token generation speed. It’s the human time spent correcting the AI’s mistakes. Cherny’s approach suggests paying the “compute tax” for a smarter model upfront eliminates the “correction tax” later. I’ve seen this play out in practice, and he’s right. A faster model that gets things wrong 30% of the time is slower than a slower model that gets things right 90% of the time, once you factor in debugging.
One shared file turns every AI mistake into a permanent lesson
Cherny also detailed how his team solves the problem of AI amnesia. Standard LLMs don’t “remember” your company’s specific coding style or architectural decisions from one session to the next. This is the single biggest pain point I’ve heard from teams trying to adopt coding agents at scale.
His solution is elegantly simple. The team maintains a single file named 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 wrote.
This practice turns the codebase into a self-correcting organism. When a human developer reviews a pull request and spots an error, they don’t just fix the code. They tag the AI to update its own instructions. “Every mistake becomes a rule,” noted Aakash Gupta, a product leader analyzing the thread.
The longer the team works together, the smarter the agent becomes. This is the kind of flywheel effect that makes me think Anthropic understands something about AI-assisted development that most others don’t. It’s not about the model being perfect out of the box. It’s about the system getting better over time through human feedback.
Slash commands and subagents automate the most tedious parts
The “vanilla” workflow one observer praised is powered by rigorous automation of repetitive tasks. Cherny uses slash commands — custom shortcuts checked into the project’s repository — to handle complex operations with a single keystroke.
He highlighted a command called /commit-push-pr, which he invokes dozens of times daily. Instead of manually typing git commands, writing a commit message, and opening a pull request, the agent handles the entire pipeline. This is the kind of automation that doesn’t sound impressive until you realize how much friction it removes from the daily workflow.
Look, I’ve been skeptical about AI coding assistants since GitHub Copilot launched. Most of them are fine for autocomplete but fall apart on anything that requires real architecture decisions. What Cherny is showing here is different. He’s not treating AI as a glorified autocomplete. He’s treating it as a workforce that he commands.
The shift from typing code to commanding agents is real. And if Cherny’s workflow is any indication, the developers who figure out how to orchestrate multiple AI agents will be the ones who ship ten times faster than everyone else.
Comments (0)
Login Log in to comment.
Be the first to comment!