The Claude Code creator’s workflow is simpler than you’d think — and that’s what makes it terrifying

Boris Cherny, the creator and head of <a href="https://video.allwinchina.org/ai-tools/claude-code/" title="Claude Code review”>Claude Code at Anthropic, shared his personal terminal setup on X last week. The engineering community has been picking it apart ever since. And honestly, I get the hype — but not for the reasons you’d expect.

This isn’t some revolutionary new paradigm. It’s not a secret sauce. It’s a guy running five Claude instances in his terminal and treating coding like a real-time strategy game. And that’s exactly why it matters.

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 might be having “their ChatGPT moment.” High praise, but let’s look at what Cherny actually does.

Five agents, one commander

Cherny doesn’t code linearly. No “write function, test, move on” loop. Instead, he runs five Claudes in parallel, 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, a third drafts documentation. He also keeps 5-10 more Claudes on claude.ai in his browser, with a “teleport” command to hand off sessions between web and local machine.

I’ve tried similar setups with other tools, and the bottleneck is always context switching. Cherny’s approach works because he’s not switching — he’s commanding. One observer on X said it “feels more like Starcraft” than traditional coding. That’s not a metaphor; it’s an accurate description of the workflow.

This fits neatly with Anthropic President Daniela Amodei’s “do more with less” strategy. While OpenAI burns cash on trillion-dollar infrastructure, Anthropic is proving that smart orchestration of existing models can yield exponential productivity gains. I’d argue the real innovation here isn’t the model — it’s the management.

The slowest model wins

Here’s the counterintuitive part: Cherny exclusively uses Opus 4.5, Anthropic’s heaviest, slowest model. “It’s the best coding model I’ve ever used,” he wrote. “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 goes against everything the industry has been optimizing for. Everyone wants lower latency, faster tokens, cheaper inference. But Cherny’s insight is brutal: the bottleneck isn’t generation speed. It’s the human time spent correcting the AI’s mistakes. Pay the “compute tax” upfront with a smarter model, and you eliminate the “correction tax” later.

I’ve fallen into this trap myself — chasing faster models only to spend more time debugging their outputs. Cherny’s approach is a reminder that productivity isn’t about raw speed; it’s about getting it right the first time.

One file to rule them all

Cherny’s team maintains a single file called CLAUDE.md in their git repository. Every time Claude does something wrong, they add it to that file. “Anytime we see Claude do something incorrectly we add it to the CLAUDE.md, so Claude knows not to do it next time.”

This turns the codebase into a self-correcting organism. 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,” noted product leader Aakash Gupta.

It’s elegant because it’s simple. No complex RAG pipelines, no fine-tuning. Just a plain text file that grows smarter over time. The longer the team works together, the better the agent gets. This is the kind of practical engineering that actually ships, unlike half the over-engineered solutions I see in the wild.

Slash commands for the tedious stuff

Cherny also uses slash commands — custom shortcuts checked into the project’s repository — to automate repetitive tasks. His most-used command is /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 boring. It’s unglamorous. But it’s the kind of automation that actually saves time. Most developers waste hours on git workflow overhead. Cherny just eliminated it.

What this means for the rest of us

The takeaway isn’t that you need to run five Claudes or use Opus 4.5. It’s that the most effective AI workflows are mundane. The magic isn’t in the model — it’s in the boring stuff: system notifications, a single config file, custom slash commands, and the discipline to treat every mistake as a learning opportunity.

Anthropic’s competitors are busy building bigger models and fancier infrastructure. Cherny is busy making his terminal sing. I know which approach I’d bet on for the next year.