Breaking

Claude Code creator says one Claude prompts other Claudes in his workflow

July 4, 2026 at 13:15 EDT

AI Agents
Software Dev & Coding

Boris Cherny, the Anthropic engineer who created Claude Code, has revealed that he no longer talks to Claude directly and instead runs a multi-layered agent setup in which one Claude issues instructions to several others. He said, "I have a Claude that prompts other Claudes. So I don't even talk to Claude," describing how his own job has shifted from writing prompts to designing and operating the coordinating loops between agents. The remark was repeated across interviews including the Big Technology Podcast and widely shared.

The creator of Claude Code · Anthropic

One Claude now prompts thousands of Claudes — while the human barely types

Boris Cherny, who built Anthropic's terminal-based coding agent, says he runs a self-recursive multi-agent setup: a meta-level Claude writes and manages prompts for other Claudes. Some mornings he oversees anywhere from hundreds to tens of thousands of AI agents — and hasn't written code by hand in eight months.

lines hand-written by Cherny in 8 months

~65%

of internal pull requests written by Claude

~4%

of public GitHub commits

~70%

engineer productivity gain (internal)

Code output, then vs now

Anthropic's reported output growth against its 2021–2025 baseline

1×

baseline

~8×

now

The self-recursive loop

Human

sets the goal, rarely types

→

Meta Claude

writes & manages prompts

→

100s–10,000s
of Claudes

investigate & implement overnight

→

PRs by morning

code + screenshots ready to review

Real-project impact

Bun's creator used Opus and Dynamic Workflows to fully rewrite Bun from Zig to Rust in 6 days — a task estimated at roughly a year for a human team.

The shift

Moving from one-off "prompt engineering" to designing autonomous agent loops
Overnight agents deliver finished work for review
Adopted at firms such as Salesforce and startups

The caution

~8× may overweight lines of code, not quality
Elite-developer cases are hard to generalize
High token cost vs engineer labor
Bottleneck migrates to review, security & culture

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.