Claude Code’s creator shared his workflow, and it’s a masterclass in AI orchestration

When the creator of the most advanced coding agent on the market casually posts his personal terminal setup, the engineering community doesn’t just read it—they dissect it, debate it, and start rebuilding their own workflows around it.

That’s exactly what happened last week when Boris Cherny, the creator and head of <a href="https://write.allwinchina.org/ai-tools/claude-code/" title="Claude Code review”>Claude Code at Anthropic, shared his workflow on X. What started as a casual thread spiraled into something the industry is calling a “watershed moment” for the company. Jeff Tang, a prominent developer, put it bluntly: “If you’re not reading the Claude Code best practices straight from its creator, you’re behind as a programmer.”

I’ve been following this thread closely, and the most striking thing is how simple the setup looks on the surface. Yet the output? One human operating with the throughput of a small engineering department. One user described the experience as “more like Starcraft than traditional coding”—a shift from typing syntax to commanding autonomous units.

Running five AI agents like a fleet commander

Cherny doesn’t code linearly. No write-test-move-next cycle. Instead, he acts like a fleet commander.

“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.”

Using iTerm2 system notifications, he manages five simultaneous work streams. One agent runs a test suite. Another refactors a legacy module. A third drafts documentation. Meanwhile, he also runs “5-10 Claudes on claude.ai” in his browser, with a “teleport” command to hand off sessions between web and local machine.

This validates a strategy Anthropic President Daniela Amodei articulated earlier this week: do more with less. While OpenAI pours money into trillion-dollar infrastructure, Anthropic is proving that better orchestration of existing models can yield exponential productivity gains. It’s a bet that feels right to me—I’ve seen too many teams throw compute at problems that needed better coordination.

The counterintuitive case for the slowest, smartest model

Here’s where Cherny goes against the grain. In an industry obsessed with latency, he exclusively uses Anthropic’s heaviest, slowest model: Opus 4.5.

“I use Opus 4.5 with thinking for everything,” he 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 building with AI. The real bottleneck isn’t token generation speed—it’s the human time spent correcting the AI’s mistakes. Cherny’s workflow suggests paying the “compute tax” upfront for a smarter model eliminates the “correction tax” later. I’ve seen this pattern play out with other tools too. The cheaper model often costs more in debugging time.

One file that turns every mistake into a permanent lesson

Cherny also addressed the elephant in the room: AI amnesia. Standard LLMs don’t remember your company’s coding style from one session to the next.

His solution? A single file named CLAUDE.md in the 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 transforms 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 Aakash Gupta, a product leader analyzing the thread. The longer the team works together, the smarter the agent becomes.

I’ve been using a similar approach with my own projects, and it’s surprisingly effective. The key is discipline—you have to actually update that file every time, which is harder than it sounds.

Slash commands and subagents automate the tedious stuff

The “vanilla” workflow that observers 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 isn’t new in concept—we’ve had git aliases for years—but the AI integration takes it to another level.

The subagent architecture is worth noting too. Cherny’s setup allows agents to spawn sub-agents for specific tasks, like running a test or checking a log, and then report back. It’s a pattern that mirrors how a good engineering lead delegates work.

Why this matters more than another model release

The developer community is calling this Anthropic’s “ChatGPT moment,” and I think that’s overblown—but not entirely wrong. What Cherny demonstrated is that the real value isn’t in the model itself, but in how you orchestrate it. That’s a lesson that applies far beyond Claude Code.

For enterprise technology leaders, the takeaway is clear: stop obsessing over which model is fastest or cheapest. Start obsessing over how you manage the workflows around it. The bottleneck isn’t the AI—it’s the human process.