In a recent revelation that has sent ripples through the software development community, Boris Cherny, the creator of Claude Code at Anthropic, shared insights into his innovative workflow that challenges traditional coding paradigms. What began as a simple showcase of his terminal setup quickly evolved into a comprehensive manifesto on the future of software engineering, capturing the attention of developers and industry insiders alike. This article delves deep into Cherny’s groundbreaking approach, exploring how it reshapes the landscape of coding and what it means for the future of technology.
Cherny’s workflow is characterized by an unconventional method of managing multiple AI agents simultaneously, a stark departure from the linear coding processes that have dominated the industry for decades. Instead of writing code in a sequential manner—where a programmer typically writes a function, tests it, and then moves on to the next task—Cherny operates more like a fleet commander in a real-time strategy game. He runs five Claude agents in parallel within his terminal, each designated to handle different aspects of the development process. This multitasking capability allows him to maximize productivity and efficiency, effectively enabling a single developer to achieve the output capacity of a small engineering team.
The mechanics of this workflow are fascinating. Cherny utilizes iTerm2 system notifications to manage his five simultaneous work streams. For instance, while one Claude agent executes a test suite, another might be tasked with refactoring a legacy module, and yet another could be drafting documentation. This orchestration of tasks not only streamlines the development process but also transforms the experience of coding itself. As one observer noted, engaging with Cherny’s setup feels more akin to playing Starcraft than traditional programming, where the focus shifts from typing syntax to commanding autonomous units.
This innovative approach aligns with the broader strategy articulated by Anthropic President Daniela Amodei, who emphasizes the importance of doing more with less. While competitors like OpenAI invest heavily in building expansive infrastructure, Anthropic demonstrates that superior orchestration of existing models can yield exponential productivity gains. Cherny’s workflow exemplifies this philosophy, showcasing how effective management of AI resources can lead to significant advancements in software development.
One of the most intriguing aspects of Cherny’s methodology is his choice of AI model. Contrary to the industry’s obsession with speed, he opts for Anthropic’s heaviest and slowest model, Opus 4.5. This decision may seem counterintuitive in a field where latency is often prioritized, but Cherny argues that the benefits outweigh the drawbacks. By using a model that requires less steering and produces fewer errors, he ultimately achieves faster outcomes. This insight is particularly valuable for enterprise technology leaders, as it highlights a critical bottleneck in modern AI development: the human time spent correcting the AI’s mistakes. Cherny’s workflow suggests that investing in a smarter model upfront can eliminate the “correction tax” that often plagues software projects.
Another significant innovation in Cherny’s workflow is the introduction of a shared file named CLAUDE.md, which serves as a repository for lessons learned from AI mistakes. Standard large language models typically lack the ability to remember specific coding styles or architectural decisions from one session to the next, leading to what Cherny refers to as “AI amnesia.” To combat this issue, his team maintains CLAUDE.md in their git repository, where they document any errors made by the AI. Whenever Claude makes a mistake, it is added to this evolving guide, ensuring that the AI learns from its past errors and avoids repeating them in the future. This practice effectively transforms the codebase into a self-correcting organism, where every mistake becomes a rule that enhances the AI’s performance over time.
The implications of this approach are profound. When a human developer reviews a pull request and identifies an error, they don’t simply fix the code; they also tag the AI to update its own instructions. This collaborative learning process fosters a dynamic relationship between human developers and AI agents, allowing both to grow and improve together. As Aakash Gupta, a product leader analyzing Cherny’s thread, aptly noted, “Every mistake becomes a rule.” The longer the team works together, the smarter the AI becomes, creating a feedback loop that continuously enhances the quality of the code produced.
Cherny’s workflow also incorporates rigorous automation of repetitive tasks through the use of slash commands—custom shortcuts that streamline complex operations into a single keystroke. One notable command he frequently employs is /commit-push-pr, which automates the process of typing git commands, writing commit messages, and opening pull requests. This level of automation not only saves time but also reduces the cognitive load on developers, allowing them to focus on higher-level problem-solving rather than getting bogged down in bureaucratic details.
Additionally, Cherny deploys specialized subagents—AI personas designed to handle specific phases of the development lifecycle. For example, he utilizes a code-simplifier to clean up architecture after the main work is completed and a verify-app agent to run end-to-end tests before any code is shipped. This division of labor among AI agents further enhances the efficiency of the development process, ensuring that each task is handled by the most suitable agent for the job.
Perhaps the most revolutionary aspect of Cherny’s workflow is the implementation of verification loops, which fundamentally change the way AI-generated code is approached. Unlike traditional AI systems that merely generate text, Claude acts as both a coder and a tester. Cherny explains that Claude tests every single change made to the codebase using the Claude Chrome extension. This extension opens a browser, conducts UI tests, and iterates until the code functions correctly and the user experience meets expectations. By giving the AI the capability to verify its own work—whether through browser automation, executing bash commands, or running test suites—Cherny claims that the quality of the final result improves by two to three times.
The significance of these verification loops cannot be overstated. They represent a paradigm shift in how developers view AI in the context of software engineering. For years, the concept of “AI coding” was largely limited to autocomplete features in text editors—a faster way to type. However, Cherny’s workflow illustrates that AI can transcend this role, functioning as an operating system for labor itself. This shift in perspective encourages developers to rethink their relationship with AI, moving away from viewing it solely as an assistant and towards recognizing it as a powerful workforce capable of driving productivity and innovation.
The reaction to Cherny’s revelations has been overwhelmingly positive, with many in the engineering community recognizing the potential of his approach to transform software development. Prominent voices in the developer community, such as Jeff Tang and Kyle McNease, have praised Cherny’s insights, suggesting that those who adopt these best practices will gain a significant advantage in their programming careers. The tools to multiply human output by a factor of five are already available; it simply requires a willingness to embrace this new mindset.
As the software development landscape continues to evolve, Cherny’s workflow serves as a beacon for what the future may hold. The integration of AI into the coding process is no longer a distant possibility; it is happening now, and those who adapt to this change will find themselves at the forefront of a new era in technology. The implications extend beyond individual productivity; they signal a broader transformation in how teams collaborate, how code is written and tested, and ultimately, how software is built.
In conclusion, Boris Cherny’s innovative workflow for Claude Code represents a significant leap forward in the realm of software engineering. By harnessing the power of AI agents, embracing a smarter model, and implementing rigorous automation and verification processes, Cherny has crafted a system that not only enhances productivity but also redefines the very nature of coding. As developers begin to adopt these practices, the potential for increased efficiency and improved code quality is immense. The future of software development is here, and it is powered by AI. Those who recognize and seize this opportunity will undoubtedly shape the next chapter of technological advancement.