When the creator of Claude Code, one of the most advanced AI coding tools to date, revealed his development process, the global developer community took notice. Boris Cherny, the head of Claude Code at Anthropic, recently shared his personal workflow, sparking widespread discussion about the future of programming and automation. What began as a glimpse into his terminal setup has evolved into a larger conversation about how developers can dramatically amplify productivity through structured AI orchestration.
Running Five AI Agents in Parallel
At the heart of Cherny’s method lies a surprising simplicity: he operates five Claude instances simultaneously in his terminal. Each instance takes on a unique role: testing, debugging, documentation, and refactoring while Cherny oversees the process like a strategist commanding a fleet. Using system notifications, he switches between sessions seamlessly, ensuring maximum parallelism without cognitive overload.
This approach enables a single developer to function with the capacity of an entire engineering team. It aligns closely with Anthropic’s philosophy of “doing more with less,” showing that intelligent orchestration of existing AI systems can rival hardware scale-ups in driving exponential efficiency.
Why “Slower” Models Win the Race
In a field dominated by the chase for faster processing, Cherny’s preference for Claude’s heaviest model, Opus 4.5, stands out. Despite its longer response time, this model reduces human correction effort thanks to higher-quality outputs and better tool integration. For enterprise developers, the lesson is clear: optimizing for accuracy and comprehension outruns raw speed in the long term. What seems like a computational slowdown becomes a strategic advantage in minimizing error correction cycles.
From Mistakes to Memory: The CLAUDE.md File
Cherny’s team also confronted a core limitation of AI assistants – context forgetfulness. Their solution: a central repository file named CLAUDE.md. Every time the AI makes an error, the correction is documented there, effectively teaching Claude to avoid repeating it. Over time, this evolving file transforms the codebase into a self-learning system that internalizes organizational standards and coding conventions.
This method converts each mistake into a permanent lesson, accelerating long-term project consistency and reducing friction between AI and human collaborators.
Automation Through Slash Commands and Subagents
A standout feature of Cherny’s workflow is extensive automation. Custom slash commands replace repetitive coding steps, allowing complex operations like committing, pushing, and opening a pull request to be executed with a single command.
Subagents further enhance the setup. Specialized AIs like verify-app and code-simplifier handle testing and cleanup phases autonomously. This modular design not only speeds up development cycles but also ensures higher-quality final outputs without manual micromanagement.
Verification Loops: The Secret to Reliability
What truly elevates Claude Code’s workflow is its built-in verification system. Rather than relying solely on human feedback, Claude continuously tests and validates its own code through browser automation and real-time execution. Cherny claims this iterative loop improves output quality “two to three times,” making the AI both coder and quality assurer in one.
This dual function addresses a common industry pain point: developers often lose significant time catching bugs that AI-generated code introduced. Claude’s ability to self-verify marks a shift from reactive debugging to proactive assurance.
The Future of AI-Powered Engineering
The broader takeaway from Cherny’s process extends beyond technical efficiency. It signals a new paradigm in how engineers conceptualize their roles. Instead of viewing AI as a tool, Cherny’s workflow positions it as a digital workforce – an autonomous collaborator capable of managing, testing, and refining its own contributions.
As adoption of such workflows grows, development could evolve from linear code-writing to strategic orchestration, where human judgment focuses on creativity and system design while AI handles execution. For forward-thinking engineers and enterprises, this represents not just evolution – but revolution- in how software gets built.
Read more such articles from our Newsletter here.
