Creator of Claude Code Reveals Workflow, Developers Are 'Losing Their Minds'

United States - Ekhbary News Agency

Creator of Claude Code Reveals Workflow, Developers Are 'Losing Their Minds'

When the creator of the world's most advanced coding agent speaks, Silicon Valley doesn't just listen — it takes notes. For the past week, the engineering community has been dissecting a thread on X (formerly Twitter) from Boris Cherny, the creator and head of Claude Code at Anthropic. What began as a casual sharing of his personal terminal setup has rapidly evolved into a viral manifesto on the future of software development, with industry insiders hailing it as a watershed moment for the AI startup.

The buzz surrounding Cherny's revelations is palpable. "If you're not reading the Claude Code best practices straight from its creator, you're behind as a programmer," wrote Jeff Tang, a prominent voice in the developer community, underscoring the perceived value of Cherny's insights. Kyle McNease, another influential industry observer, amplified the sentiment, declaring that with Cherny's "game-changing updates," Anthropic is "on fire," potentially experiencing "their ChatGPT moment."

The core of the excitement lies in a fascinating paradox: Cherny's workflow is remarkably simple in its conceptualization, yet it empowers a single human to operate with the output capacity typically associated with a small engineering department. One user on X, after implementing Cherny's setup, described the experience as feeling "more like Starcraft" than traditional coding. This analogy highlights a fundamental shift from the manual, syntax-driven process of coding to a more strategic, command-oriented interaction with autonomous AI agents.

This article delves into the workflow that is reshaping how software gets built, directly from its architect. It explores how running multiple AI agents simultaneously transforms the coding process into something akin to a real-time strategy game.

Orchestrating AI Agents for Maximum Output

The most striking revelation from Cherny's disclosure is his departure from linear coding practices. The traditional "inner loop" of development involves a programmer writing a function, testing it, and then proceeding to the next. Cherny, however, operates more like a conductor or a fleet commander, managing multiple AI assistants concurrently.

"I run 5 Claudes in parallel in my terminal," Cherny revealed. "I number my tabs 1-5, and use system notifications to know when a Claude needs input." By leveraging iTerm2 system notifications, Cherny efficiently manages five simultaneous work streams. For instance, while one AI agent is executing a test suite, another might be refactoring a legacy module, and a third could be drafting documentation. Complementing this, he also runs "5-10 Claudes on claude.ai" in his browser, utilizing a "teleport" command to seamlessly transfer sessions between the web interface and his local machine.

This approach strongly validates the "do more with less" strategy recently articulated by Anthropic President Daniela Amodei. In stark contrast to competitors like OpenAI, who are pursuing massive infrastructure build-outs, Anthropic is demonstrating that superior orchestration and intelligent utilization of existing AI models can yield exponential productivity gains. This highlights a shift in competitive strategy within the AI landscape, focusing on efficiency and synergy rather than solely on raw computational power.

The Counterintuitive Choice: The Slowest, Smartest Model

In a move that might seem counterintuitive for an industry obsessed with speed and low latency, Cherny disclosed his exclusive use of Anthropic's most powerful, albeit slowest, model: Opus 4.5. "I use Opus 4.5 with thinking for everything," Cherny 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 insight is critical for enterprise technology leaders. The primary bottleneck in modern AI development is often not the speed of token generation but the human time required to correct AI errors. Cherny's workflow suggests that accepting a higher "compute tax" for a more capable, intelligent model upfront can significantly reduce the subsequent "correction tax," leading to greater overall efficiency and faster time-to-market.

Persistent Learning: A Shared File for AI Memory

Cherny also detailed his team's innovative solution to the problem of AI "amnesia." Standard large language models often lack persistent memory of a company's specific coding styles or architectural decisions across different sessions.

To overcome this, Cherny's team maintains a single, central file named CLAUDE.md within their 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 practice effectively transforms the codebase into a self-correcting, learning organism. When a human developer reviews a pull request and identifies an error, they not only fix the code but also update the CLAUDE.md file, instructing the AI on how to avoid similar mistakes in the future. As product leader Aakash Gupta noted while analyzing the thread, "Every mistake becomes a rule." The longer the team collaborates with the AI using this method, the more intelligent and aligned the agent becomes.

Automation Through Slash Commands and Sub-Agents

The highly efficient "vanilla" workflow that has drawn praise is powered by rigorous automation of repetitive development tasks. Cherny utilizes slash commands and sub-agent concepts to automate time-consuming activities such as generating test cases, creating pull requests, and handling minor bug fixes. This allows human developers to dedicate their cognitive resources to more complex problem-solving and creative aspects of software engineering, ultimately pushing the boundaries of what's possible in AI-driven development.

Ekhbary News Agency