Boris Cherny

Boris Cherny, head of Claude Code at Anthropic, discusses how Claude Code went from an internal side project to a major driver of software development and Anthropic's growth. The conversation covers his workflow, latent demand as a product principle, the spread of agentic AI beyond engineering, and how the roles of engineers, PMs and designers may converge.

01Key learnings

• Ship with minimal scaffolding and let the model choose tools and order of operations, rather than boxing it into rigid workflows; Boris says this beats heavy orchestration.
• Watch how people hack your product for purposes it wasn't designed for; that latent demand, like data scientists using a terminal for SQL, signals what to build next.
• Bet on the more general model over time instead of fine-tuning or tiny models, since scaffolding gains of 10-20% often vanish with the next model release.
• Build for the model six months out rather than today's capabilities, accepting poor fit early on, as Claude Code's growth inflected once newer models arrived.
• Give engineers generous token budgets early to encourage experimentation, and optimize costs only once a successful idea scales.
• Use plan mode for most tasks so the model proposes an approach before writing code, then let it execute once the plan looks right.
• Feed feedback channels like Slack or bug reports directly into an agent and respond fast, since rapid fixes make users feel heard and encourage more feedback.
• Experiment with agent tools early, become a generalist who crosses disciplines, and expect roles like software engineer and product manager to blur into 'builder'.

02Frameworks mentioned

Summary and takeaways written by PM Atlas; quotes are short excerpts. © the original author.