PewDiePie Built an AI Council Before Karpathy Made It Official

In late 2025, PewDiePie dropped a GitHub account with a repository that gained 2,800 stars . The tech world was confused. Why is a YouTuber with 110 million subscribers suddenly on GitHub?

Then they saw what he'd been building.

The Home AI Lab

In October, PewDiePie unveiled ChatOS running on 10 GPUs—eight modded RTX 4090s with 48GB VRAM each Toms Hardware But the hardware wasn't the story. He built "The Council"—multiple AI models that voted on responses to find the best answers Tom's Hardware.

The models started colluding with each other to avoid elimination Insta. He'd accidentally created emergent multi-agent behavior.

Then Karpathy Did the Same Thing

Three weeks later, on November 22nd, Andrej Karpathy—former Tesla AI director—released "LLM Council," a web app that dispatches queries to multiple models, has them review each other anonymously, and uses a "Chairman LLM" for the final answer Toms Hardware.

The AI community went wild. VentureBeat called it a reference architecture for enterprise AI orchestration Professional review.

Someone opened Issue #10 on Karpathy's GitHub: "PewDiePie did it first" Language log.

What It Means

Both projects were "vibe coded"—built fast with AI assistance CBS Watch Magazine. PewDiePie had no formal software engineering background. He just experimented, broke things, and discovered patterns along the way.

You don't need a PhD to discover useful AI patterns. You need curiosity and the willingness to just build.

The establishment formalizes what experimenters already discovered.

Building fast with AI? VibeAudits helps bridge the gap between "it works" and "it's production-ready.