- On February 23, 2026, Boris Cherny—leader of Claude Code at Anthropic—shared three core principles applied to his entire team.
- Principle 1: “Better than doing it yourself is letting Claude do it.” He emphasized that Claude can automate the vast majority of technical work.
- Principle 2: “Underfund everything a little bit”—deliberately keeping teams small and resources limited to force employees to “Claude-ify” their workflows.
- However, Cherny advises CTOs not to cut costs too early; initially, engineers should be given “as many tokens as possible.”
- Tokens—the billing unit for AI models—are becoming a major concern as CFOs are surprised by the surging AI costs per engineer.
- Cherny believes cost optimization should only occur once a product has proven its value and consumes significant tokens; one should not optimize too early.
- Principle 3: Prioritize speed. He stated that when a team is very small, speed is the only competitive advantage in a crowded coding market.
- Recently, Anthropic and OpenAI released programming tool updates only minutes apart.
- Previously, the Claude team used Claude itself to build “Claude Cowork”—a non-technical Agent—in just 10 days.
- Cherny concluded: to accelerate, let Claude do more work instead of traditional hiring.
Conclusion: Boris Cherny’s three principles at Anthropic revolve around maximizing AI over human labor: automating with Claude, intentional “underfunding” to force AI dependency, and prioritizing deployment speed. Although tokens can significantly increase costs per engineer, he advises against premature optimization. In a race where updates between Anthropic and OpenAI happen within minutes, speed and automation become vital advantages.
