AI evaluation ecosystem model puts LLM agents in institutional roles

A thread about new work modeling and simulating the AI evaluation ecosystem says LLM agents play strategy teams, investment committees and a regulator. It says markets, benchmark scoring and incidents follow fixed rules, so each channel can be changed one at a time. Agents see public scores and headlines, while strategy and beliefs remain private; true capability and user satisfaction never enter a prompt.