Updated 13 hours ago
Anthropic Claude Code Creator Manages Tens of Thousands of AI Agents at Once

Claude Code at Scale

Anthropic Claude Code Creator Manages Tens of Thousands of AI Agents at Once

Boris Cherny, creator of Claude Code at Anthropic, has not written a line of code by hand in eight months. Instead, he orchestrates fleets of AI agents — sometimes tens of thousands at once — that write, review, and even conceive new features autonomously.

Eight Months Without Writing a Line of Code

Boris Cherny, the creator and head of Claude Code at Anthropic, has not written a line of code by hand in eight months. Speaking at Fortune Brainstorm Tech 2026 in Aspen on Monday, Cherny revealed that he now orchestrates vast fleets of AI agents — sometimes hundreds, sometimes tens of thousands — that write, review, and even conceive new features autonomously, according to Fortune.

"This morning I was managing maybe a few hundred. Some days it is thousands, or tens of thousands," Cherny said, per.1 "It is quite exciting, because to me it means I can do much more, and the role of a builder is just totally changing."

From Terminal Window to Multi‑Agent System

A year and a half ago, Claude Code was a single instance running in one terminal window. Today, it is a self‑prompting, multi‑agent system where one Claude delegates work to others — and increasingly, the human is not even in the prompting loop.

"Fast‑forward to today, it looks very different. You have a Claude Code, but it has subagents that are other Claudes," Cherny explained, as reported by Fortune. "It is actually another Claude that does the prompting."

The Tool That Writes Itself

All code at Anthropic is now written by Claude Code. Since the beginning of 2026, code output has increased 8x. Claude Code fully writes itself, including performing its own security reviews, and actively scans GitHub and X to decide what to build next.

"We are starting to get to the point where it has ideas. It is looking at GitHub, it is looking at X; it is figuring out, What should I build next?" Cherny said, Fortune reported. Often, when he wakes up, Claude has already acted on new ideas autonomously.

  • All code at Anthropic Now written by Claude Code — not just assisted, but authored
  • Code output Increased 8x since January 2026
  • Self‑writing Claude Code writes its own codebase, including security reviews
  • Autonomous ideation Scans GitHub and X to decide what features to build next
  • Multi‑agent delegation Higher‑level Claudes prompt lower‑level ones — human out of the loop

The Printing Press Moment

Cherny compared the speedup in coding to the impact of the printing press, which dramatically lowered the cost of book production and eventually sparked the Renaissance and Industrial Revolution. Lowering the barrier to software creation could trigger a similar wave of innovation. "I think this might be the first product that actually just takes off," Cherny said, per.1 Claude Code has become the most‑used AI coding tool among 15,000 surveyed developers and now accounts for roughly 4% of all GitHub commits, SERPsculpt reports, making it one of the fastest‑growing developer products in history.

The Recursive Self‑Improvement Problem

Cherny was also candid about the risks. When asked about recursive self‑improvement — where AI systems autonomously design and build better versions of themselves — he called it "one of the big risks for AI," Fortune reported. Anthropic published a blog post the previous week titled "When AI Builds Itself," outlining how AI systems increasingly develop future AI models. For builders, the implications are twofold: AI tools are becoming dramatically more capable, but the same systems that write code are beginning to write themselves — and that threshold is only starting to be grappled with.

Sources

  1. 1.Fortune(fortune.com)
  2. 2.SERPsculpt(serpsculpt.com)

Share this article

PostShare

More on This Story

Related News