Anthropic's Boris Cherny manages tens of thousands of AI agents daily as Claude Code writes itself and Anthropic's own code

Cherny, creator and head of Claude Code at Anthropic, stated he manages anywhere from a few hundred to tens of thousands of AI agents on different days. The setup now features Claude Code with subagents that are other Claude instances, where one Claude prompts another Claude instead of a human user doing the prompting.

Claude Code has led to an 8X increase in the amount of code written at Anthropic since the beginning of the year. The system is fully writing itself and performing its own security review, and has begun generating its own ideas by examining GitHub and other sources to determine what to build next.

Cherny compared the speedup in coding to the printing press (developed around 1440), saying it lowers barriers to software creation and could unlock a wave of innovation. Anthropic published a blog post titled 'When AI Builds Itself,' describing how it uses AI systems to develop future AI models, with the potential for recursive self-improvement—where an AI system autonomously designs and improves its own successors. Cherny acknowledged that recursive self-improvement is 'one of the big risks for AI.'