When the creator of the world’s most advanced coding agent speaks, Silicon Valley doesn’t just listen – it takes notes.
Over the past week, the engineering community has been analyzing a thread on X With Boris Czernycreator and boss Claude Koda On Anthropic. What started as an accidental sharing of the configuration of his personal terminal turned into a viral manifesto about the future of software development, and industry insiders are calling it a watershed moment for the startup.
“If you don’t read Claude Code best practices straight from its creator, you’re behind as a programmer,” he wrote Jeff Tanga prominent voice in the developer community. Kyle McNeaseanother industry observer, went further and declared that thanks to “game-changing updates,” Cherny Anthropic is “on fire,” potentially waiting for a “ChatGPT moment.”
The excitement comes from a paradox: Cherny’s workflow is surprisingly elementary, yet allows a single person to operate with the efficiency of a diminutive engineering department. As one user on X noted after implementing the Cherny setup, the experience “it’s more like Starcraft” than traditional coding – moving from the syntax of writing to commanding autonomous units.
Here’s a workflow analysis that’s changing the way you build software, straight from the architect himself.
How running five AI agents simultaneously turns coding into a real-time strategy game
The most striking discovery from Cherny’s disclosure is that he does not encode in a linear fashion. In traditionalinner loop” in development, a programmer writes a function, tests it, and moves on to the next one. Cherny, however, serves as fleet commander.
“I’m running 5 Claudes in parallel in my terminal,” Cherny wrote. “I number my cards 1 to 5 and use system notifications to let me know when Claude needs help.”
Using iTerm2 system notifications, Cherny effectively manages five simultaneous work streams. While one agent runs the test suite, another refactors the legacy module, and the third prepares documentation. He also hosts the program “5-10 Claudes”. claude.ai” in your browser using the “teleport” command to transfer the session between the network and your local computer.
This confirms “do more with less“a strategy presented by Anthropic CEO Daniela Amodei earlier this week. While competitors like OpenAI pursue trillion-dollar infrastructure builds, Anthropic proves that perfect orchestration of existing models can deliver exponential productivity gains.
A counterintuitive case of choosing the slowest and smartest model
In a surprising move for a lag-obsessed industry, Cherny revealed that he only uses the heaviest and slowest Anthropic model: Opus 4.5.
“I use Opus 4.5 and think about everything” – Cherny explained. “It’s the best encoding model I’ve ever used, and even though it’s larger and slower than Sonnet because it needs less control and is better at tooling, it’s almost always ultimately faster than using a smaller model.”
For enterprise technology leaders, this is a key observation. The bottleneck in the development of modern artificial intelligence is not the speed of token generation; is human time spent fixing AI errors. Cherny’s workflow suggests that paying the “assessed tax” upfront for a smarter model eliminates the “adjustment tax” later.
One shared file turns every AI mistake into a lasting lesson
Cherny also detailed how his team is solving the AI amnesia problem. Standard models for large languages do not “remember” a company’s specific coding style or architectural decisions between sessions.
To solve this problem, Cherny’s team maintains a single file called CLAUDE.md in their git repository. “Every time we see Claude doing something wrong, we add it to CLAUDE.md so Claude knows not to do it next time,” he wrote.
This practice transforms the codebase into a self-correcting organism. When a developer is reviewing a pull request and notices a bug, they don’t just fix the code; mark the AI to update its own instructions. “Every mistake becomes a rule”, he noted Akash Guptaproduct leader analyzing the thread. The longer a team works together, the smarter the agent becomes.
Slash commands and subagents automate the most tedious parts of programming
The “vanilla” workflow that one observer praised relies on tough automation of repetitive tasks. Cherny uses slash commands – custom shortcuts marked in the project repository – to handle complicated operations with a single keystroke.
He highlighted the command named /commit-push-prthat is referenced dozens of times a day. Instead of manually typing git commands, writing a commit message, and opening a pull request, the agent handles the version control bureaucracy on its own.
Cherny also deploys subagents — specialized AI people — to handle specific phases of the software lifecycle. It uses a code simplification tool to spotless up the architecture after major work is completed, and an application verification agent to perform end-to-end testing before anything is shipped.
Why verification loops are the true unlocker of AI-generated code
If there’s one reason Claude Code reportedly hit $1 billion in annual recurring revenue so rapid, it’s probably a verification loop. Artificial intelligence is not just a text generator; this is a tester.
“Claude tests every change I make to claude.ai/code using the Claude Chrome extension,” Cherny wrote. “He opens the browser, tests the UI, and iterates until the code works and the UX is good.”
He claims that allowing AI to verify its own work – whether by automating the browser, running bash commands, or executing test suites – improves the quality of the end result by a factor of “2-3.” An agent doesn’t just write code; this is proof that the code works.
What Cherny’s workflows signal the future of software engineering
The reaction to Cherny’s thread suggests a key shift in the way developers think about their craft. For years, “AI coding” meant the autocomplete feature of a word processor – a faster way to write. Cherny has shown that it can now function as an operating system for labor itself.
“Read this if you’re already an engineer… and want more power” Jeff Tang summed up in X.
The tools to quintuple human productivity are already available. They just require the willingness to stop thinking of AI as an assistant and start treating it like a workforce. Developers who are the first to make this mental leap will not only be more productive. They will be playing a completely different game and everyone else will still be writing.