The Claude Code creator’s workflow is a masterclass in treating AI like a fleet, not a tool

When the person who built the most hyped coding agent in the industry casually drops his personal workflow, you’d expect some arcane, multi-monitor, 40-keybind setup that takes months to learn. Instead, Boris Cherny, creator and head of <a href="https://wpkvip.com/ai-tools/claude-code/" title="Claude Code review”>Claude Code at Anthropic, posted something that looks more like a real-time strategy game than a development environment.

And the engineering community has been dissecting it for a week straight.

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.”

That’s a lot of hype. But the interesting part is that the actual substance backs it up.

Five agents, one commander

Cherny doesn’t code in a straight line. No inner loop of write-test-next. He runs five Claude instances in parallel in 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 Claudes running on claude.ai in his browser, using a “teleport” command to hand off sessions between web and local machine.

This is the kind of workflow that makes you realize how much of traditional development is just waiting. Waiting for builds, waiting for tests, waiting for your brain to context-switch. Cherny’s approach treats those waits as opportunities to keep other agents busy.

One user on X described the experience after implementing this setup as “feeling more like Starcraft” than coding. That’s not just a clever analogy — it’s a fundamentally different way of thinking about software production. You’re not typing syntax; you’re commanding autonomous units.

This validates the “do more with less” strategy Anthropic President Daniela Amodei talked about recently. While OpenAI throws money at trillion-dollar infrastructure, Anthropic is betting that better orchestration of existing models yields exponential gains. I’d say that bet is looking pretty good.

The slow model paradox

Here’s the counterintuitive part: Cherny exclusively uses Opus 4.5, Anthropic’s heaviest, slowest model. In an industry obsessed with latency, he’s choosing the opposite.

“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 bottleneck isn’t token generation speed — it’s the human time spent correcting mistakes. Pay the “compute tax” for a smarter model upfront, and you eliminate the “correction tax” later. I’ve seen teams burn weeks on fixing bad outputs from cheaper models, and it’s never worth it.

The CLAUDE.md that learns

Cherny’s team maintains a single file called CLAUDE.md in their git repository. Every time the AI does something wrong — uses the wrong coding style, makes an architectural misstep — 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,” he explained.

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. Aakash Gupta, a product leader analyzing the thread, called it “every mistake becomes a rule.” The longer the team works together, the smarter the agent gets.

This is the kind of simple, elegant solution that makes you wonder why more teams aren’t doing it. It’s almost too obvious in retrospect.

Slash commands and the death of busywork

The “vanilla” workflow that observers praised is powered by rigorous automation of repetitive tasks. Cherny uses custom slash commands checked into the project’s repository — shortcuts that handle complex operations with a single keystroke.

One example: a command called /commit-push-pr that 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 thing that sounds minor until you realize how much mental overhead it eliminates. Every time you avoid context-switching to a terminal to type git commands, you preserve focus for actual problem-solving.

What this means for the rest of us

Cherny’s workflow is notable not because it’s revolutionary in a technical sense, but because it’s practical. Five parallel agents, a single configuration file, and a preference for the slowest, smartest model. Any team can implement this today.

The real takeaway isn’t about Claude Code specifically — it’s about the shift in how we think about AI assistance. The best workflows treat AI not as a tool you use, but as a workforce you command. And the humans who figure that out first are going to have an unfair advantage.