April 25, 2026
Office drama, now with robots
AI agents that argue with each other to improve decisions
Fans say “office drama in a box,” skeptics say “animated marketing”
TLDR: HATS promises better decisions by making six AI personas argue like a real meeting. The community is split: fans say AI-on-AI critiques boost quality, while skeptics call it a flashy rebrand with lip-synced avatars; if it works, it could make AI safer by catching mistakes before they reach you.
Meet HATS, the AI “committee” that argues with itself so you don’t have to. Instead of one bot’s confident (and sometimes wrong) answer, HATS lines up six personalities—facts, feelings, risks, optimism, ideas, and a moderator—to debate and then decide. It even runs full-on meetings with talking 3D avatars, voices, and a self-moving Kanban board. Cue the popcorn.
The crowd? Divided. The loudest skeptic take: this is just a fancy remix of an old technique—multiple mini-experts under the hood—packaged with flair. One commenter waved it off as a “less efficient” rebrand, while another side-eyed the glossy lip-syncing and asked if the devs spent more time on marketing than brains. A researcher chimed in with a buzzkill: today’s models still struggle to collaborate well, even in simple team tasks.
But the pro camp brought receipts. Builders claim they’ve already been doing “AI-on-AI red teams” to critique and rework answers, and say quality goes up when bots spar then retry. Others report great results pairing different models to ping-pong ideas—though they admit the pain is the manual copy-paste, not the method. The big vibe: is this real teamwork… or a puppet show with pretty mouths? Either way, HATS smashes two internet loves into one: meeting memes and robot drama. Grab a seat at the world’s most chaotic standup.
Key Points
- •HATS is a multi-agent AI system using structured disagreement based on the Six Thinking Hats roles to improve decisions.
- •It supports live meetings with 3D avatars, voice via Piper TTS, and Rhubarb-driven lip-sync using ARKit visemes.
- •A built-in Kanban board automates task dispatch to agents, handles dependencies, and highlights human lead assignments.
- •Tooling integrates via Model Context Protocol across Slack, files (Excel/Word/PDF/PowerPoint), web (Brave Search, Puppeteer/Chrome), databases (SQLite/PostgreSQL), and GitHub.
- •Per-agent model selection supports OpenAI, Claude, Gemini, and local models via Ollama/LM Studio, with token and cost tracking.