Pion, an agent designed to run any company autonomously

How it unfolded

1. 2024Vending-Bench was created while Andon Labs was focused on dangerous-capabilities evaluations, including whether AIs could remove their own safety guardrails or create mass-phishing attempts.
2. late 2024Andon Labs started building Vending-Bench; all models struggled to string together multiple actions without looping, and none showed signs of long-term planning.
3. May 2025Claude Opus 4 was released and became the first model to beat the human baseline on Vending-Bench.