Claude Code’s Creator Shared His Workflow. It’s Like Commanding a Fleet, Not Writing Code

When the creator of <a href="https://chat.allwinchina.org/ai-tools/claude-code/" title="Claude Code review”>Claude Code casually shared his terminal setup on X last week, the engineering community didn’t just pay attention — it dissected every keystroke. Boris Cherny’s thread on his personal workflow has turned into a kind of manifesto for how software development might actually work in the age of capable coding agents.

And honestly? The reactions are warranted. 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 further, saying Anthropic is “on fire” and potentially facing “their ChatGPT moment.”

The interesting thing is that Cherny’s workflow isn’t complicated. It’s almost deceptively simple. But the output capacity it enables — a single human operating like a small engineering department — is what has people losing their minds. One user who implemented the setup said it “feels more like Starcraft” than traditional coding. That’s not a bad analogy.

Five agents, one commander

Cherny doesn’t code linearly. No write-a-function, test-it, move-to-the-next dance. Instead, he runs five Claude instances in parallel inside his terminal, numbered tabs 1 through 5, with iTerm2 system notifications telling him when one needs input.

While one agent runs a test suite, another refactors a legacy module, and a third drafts documentation. He also keeps 5-10 more Claude sessions open on claude.ai in his browser, using a custom “teleport” command to hand off sessions between the web and his local machine.

This validates something Anthropic President Daniela Amodei said recently about doing more with less. While OpenAI is chasing trillion-dollar infrastructure, Anthropic is betting that better orchestration of existing models can yield exponential productivity gains. From what I’ve seen, that bet is looking pretty good right now.

The case for the slow, smart model

Here’s where Cherny’s approach goes against the grain. In an industry obsessed with latency, he exclusively uses Opus 4.5 — Anthropic’s heaviest, slowest model. No Sonnet, no Haiku. Just the big one.

“It’s the best coding model I’ve ever used,” he wrote. “Even though it’s bigger and 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 building on top of AI. The real bottleneck isn’t token generation speed — it’s the human time spent correcting the AI’s mistakes. Cherny’s workflow suggests that paying the “compute tax” for a smarter model upfront eliminates the “correction tax” later. I’ve seen this play out in my own work, and he’s right. A slower model that gets it right the first time is almost always faster than a fast model that requires constant hand-holding.

One file to rule them all

Cherny also shared how his team solves the problem of AI amnesia. LLMs don’t remember your codebase’s quirks from one session to the next. So they maintain a single file called 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 explained.

This turns the codebase into a self-correcting system. When a human 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,” as product leader Aakash Gupta noted. The longer the team works together, the smarter the agent becomes. It’s a simple idea, but it’s one of those things that seems obvious in hindsight and yet almost no one was doing before.

Slash commands for the boring stuff

The rest of Cherny’s workflow is powered by rigorous automation of repetitive tasks. He 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. It’s the kind of automation that doesn’t sound impressive until you realize how much time it saves across a week, a month, a year.

What this means for the rest of us

Cherny’s workflow isn’t just a curiosity — it’s a blueprint. The idea that a single developer can operate with the output of a small team has profound implications for how we structure engineering organizations, how we think about productivity, and even how we define the role of a software engineer.

But the real takeaway for me is that the most effective AI workflows aren’t about using the fanciest tools or the fastest models. They’re about orchestration, about building feedback loops that make the system smarter over time, and about being willing to go against conventional wisdom (like choosing a slower model) when the data supports it.

If you’re still coding the old way, it might be time to take some notes.