When the creator of one of the world’s most advanced coding agents speaks, Silicon Valley doesn’t just listen — it takes notes. Boris Cherny, the head of Claude Code at Anthropic, recently shared his personal workflow on X, sparking a wave of excitement and analysis across the developer community. What started as a casual look into his terminal setup has evolved into a viral manifesto on the future of software development, with many calling it a watershed moment for the company.
How running five AI agents at once turns coding into a real-time strategy game
Cherny’s approach defies traditional software development patterns. Rather than working through a linear inner loop — writing, testing, and iterating — he commands a fleet of AI agents in parallel. Using iTerm2 notifications and tabbed interfaces, he runs five instances of Claude simultaneously, each handling distinct tasks like testing, refactoring, or documentation drafting. This setup allows him to manage multiple streams of work as if commanding autonomous units — a shift that many developers describe as feeling more like Starcraft than typical coding.
The counterintuitive case for choosing the slowest, smartest model
One surprising aspect of Cherny’s workflow is his exclusive use of Anthropic’s largest and slowest model, Opus 4.5. While many in the industry prioritize speed and efficiency, Cherny argues that a smarter, more capable model reduces the need for constant human correction, ultimately saving time. By using the most powerful model for all tasks, he minimizes the “correction tax” that comes with less intelligent AI, making the process faster in the long run.
Slash commands and verification loops automate the future of coding
Another key element of Cherny’s method is the use of automation to eliminate mundane tasks. He employs slash commands like /commit-push-pr to streamline version control, and deploys specialized subagents for code simplification and end-to-end testing. Crucially, he uses a shared file, CLAUDE.md, to ensure that every mistake becomes a permanent lesson, enabling the AI to learn from past errors. This creates a self-improving system where Claude gets smarter over time, and the developer’s role evolves from code writer to orchestration commander.
Cherny’s workflow signals a fundamental shift in how developers interact with AI — from using it as a typing assistant to treating it as a workforce. As Anthropic’s Claude Code reaches $1 billion in annual recurring revenue, the industry is watching closely to see how this new paradigm reshapes the future of software engineering.
