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 from Boris Cherny, creator and head of cloud code at Anthropic. What started as a casual sharing of their personal terminal setups has turned into a viral manifesto on the future of software development, with industry insiders calling it a watershed moment for the startup.
"If you are not reading cloud code best practices directly from its creator, then as a programmer you are behind," Written by Jeff Tang, a leading voice in the developer community. Another industry observer, Kyle McNeese, added: "Game-changing update," is anthropological "On fire," potentially facing "Their chatgpt moment."
The excitement stems from a paradox: Cherney’s workflow is surprisingly simple, yet it allows a single human being to do work with the output capacity of a small engineering department. As one user noted on X after implementing Cherney’s setup, the experience "feels more like starcraft" Compared to traditional coding – the shift from typing syntax to commanding autonomous units.
Here’s a breakdown of the workflow that’s reshaping the way architects themselves build software.
How running five AI agents simultaneously turns coding into a real-time strategy game
The most shocking revelation from Cherney’s revelations is that he does not code in a linear manner. in traditional "inner loop" In the course of development, a programmer writes a function, tests it, and moves on to the next. However, Cherny serves as a fleet commander.
"I run 5 clouds in parallel in my terminal," Cherney wrote. "I number my tabs 1-5, and use system notifications to know when the cloud needs input."
Using iTerm2 system notifications, Cherny effectively manages five work streams simultaneously. While one agent runs a test suite, another refactors a legacy module, and a third drafts documentation. he also runs "5-10 Claude on claude.ai" In your browser, using a "teleport" The command to close the session between the web and its local machine.
This confirms "do more with less" The strategy was expressed by Anthropic President Daniela Amodei earlier this week. While competitors like OpenAI are attempting to build trillion-dollar infrastructure, Anthropic is proving that better orchestration of existing models can yield rapid productivity gains.
The counter-intuitive case for choosing the slowest, smartest model
In a surprising move for an industry plagued by latency, Cherney revealed that he exclusively uses Anthropic’s heaviest, slowest model: the Opus 4.5.
"I use Opus 4.5 to think about everything," Cherney explained. "This is the best coding model I’ve used so far, and even though it’s bigger and slower than Sonnet, because you have to run it less and the tool is better to use, in the end it’s almost always faster than using a smaller model."
For enterprise technology leaders, this is an important insight. The barrier to modern AI development is not the speed of token generation; This is human time spent correcting the AI’s mistakes. Cherney’s workflow suggests making payments "calculate tax" Eliminates it in advance for a smart model "make improvements" Later
A shared file turns every AI mistake into a permanent lesson
Cherney also explained how his team solves the problem of AI forgetting. Standard large language models don’t do this "Memorization" Company-specific coding styles or architectural decisions from one session to the next.
To address this, Cherney’s team maintains a single file called CLAUDE.md in its Git repository. "Whenever we see Cloud doing something wrong we add it to CLAUDE.md, so that Cloud knows not to do it next time," He has written.
This practice turns the codebase into a self-improving organism. When a human developer reviews a pull request and sees an error, they don’t just fix the code; They tag the AI to update its own instructions. "Every mistake becomes a rule," said Akash Gupta, a product leader who analyzed the thread. The longer the team works together, the smarter the agent becomes.
Slash commands and subagents automate the most difficult parts of development
"vanilla" The workflow praised by a supervisor is driven by rigorous automation of repetitive tasks. Cherney uses slash commands – custom shortcuts checked out of the project’s repository – to handle complex operations with a single keystroke.
He highlighted a command called /commit-push-prWhich he invokes dozens of times every day. Instead of manually typing git commands, writing a commit message, and opening a pull request, the agent handles the bureaucracy of version control autonomously.
Cherney also deploys sub-agents – specialized AI personalities – to handle specific stages of the development lifecycle. He uses a code-simplifier to clean up the architecture after the main work is complete and a validation-app agent to run end-to-end tests before shipping anything.
Why are verification loops the real unlock for AI-generated code?
If there’s one reason why Cloud Code has reportedly hit $1 billion in annual recurring revenue so quickly, it’s probably the validation loop. AI is not just a text generator; This is a tester.
"Claude tests every change coming to claude.ai/code using a Chrome extension," Cherney wrote. "This opens a browser, tests the UI, and repeats until the code works and the UX looks good."
They argue that giving AI a way to verify its work – whether through browser automation, running bash commands, or executing test suites – improves the quality of the end result. "2-3x." The agent doesn’t just write code; This proves that the code works.
What does Cherney’s workflow indicate about the future of software engineering
The responses to Cherney’s thread reveal how developers think about their craft. for years, "ai coding" This was meant to have an autocomplete function in a text editor – a faster way to type. Cherney has demonstrated that it can now function as an operating system for labor.
"Read this if you are already an engineer… and want more power," Jeff Tang presented the summary on X.
The tools to increase human production fivefold are already here. They just need a willingness to stop thinking of AI as an assistant and treat it as a workforce. Programmers who make the mental leap earlier will not only be more productive. They’ll be playing a completely different game – and everyone else will still be typing.
<a href