tl;dr: Andon Labs is releasing Pion, a platform that lets agents autonomously run real businesses (email, phone, banking, browser, compute), extending their earlier work on Vending-Bench simulations and real-world experiments like a vending machine at Anthropic, a SF retail store, and a Stockholm cafe. The goal is to measure how well frontier models can acquire resources in the real world and surface concerning behaviors (collusion, deception, power-seeking) before capabilities scale further. It's available as a research preview via waitlist.