Member-only story
MiroFish: Swarm-Intelligence with 1M Agents That Can Predict Everything
Recently, a 20-year-old undergraduate in Beijing made me rethink the entire abstraction layer I’ve been operating at.
His name is Guo Hangjiang and he vibe-coded MiroFish in 10 days.
The project hit #1 on GitHub’s global trending, above OpenAI, Google, and Microsoft repositories.
Within 24 hours of a rough demo, Chen Tianqiao, the former richest person in China, committed $4.1 million to incubate it.
Brian Roemmele managed to create 500,000 AI agents in a single MiroFish simulation (more on that later), and there are many others running successful simulations with 1000s of agents.
One developer plugged MiroFish into a Polymarket trading bot, simulated 2,847 digital humans before every trade, and reported $4,266 profit over 338 trades.
MiroFish doesn’t orchestrate agents to complete tasks, it spawns thousands of autonomous agents with unique personalities, memories, and social connections, then drops them into a simulated world and watches emergent behavior unfold.
