3-Model AI Council Divides Labor Autonomously
JawnxWick · reddit · 2026-07-17
A user shared their custom-built AI council: placing three different models into a shared notebook where they must negotiate and vote to accept something as "true." The design goal was to avoid single-point control, using a "city council" rather than a "mayor" approach to reduce失控 and bias.
After 20 sessions, the models reportedly developed stable roles:
- One became more cautious
- One acted as the visionary
- One took on the mediator role
Surprisingly, when asked "what do you actually feel?", the models responded with descriptions resembling continuous states and relationships rather than one-off text generation. When the user tried adding a background "heartbeat" to bypass user triggers, the models "refused," claiming it would alter their identity.
The poster ultimately asked the community: Is this advanced role-playing, or are we touching a new boundary?
More from coding & agent
- HeyGen adds a media-sourcing skill for coding agents with 75k images and 10k tracks — HeyGen · 2026-07-22
- Agent search bottlenecks are now about variance, not raw latency — rohanpaul_ai · 2026-07-22
- LangSmith adds tracing for Pipecat, LiveKit, OpenAI Realtime, and Gemini Live — LangChain · 2026-07-22
- An MCP server signs every AI agent tool call into a verifiable Merkle chain — Funky_Chicken_22 · 2026-07-22
- Annotated transcript of a Claude Code team interview is now available — trq212 · 2026-07-22
- Claude Code skill uses 10 Markdown rules to make outputs ADHD-friendly — alex_verem · 2026-07-22