An AI safety research startup that deploys frontier AI models in real autonomous businesses and builds benchmarks evaluating AI agent behavior.
Andon Labs is an AI safety research startup, founded in 2024 and backed by the Y Combinator accelerator. The company studies and deploys frontier AI models in real, autonomous organizations to test whether they can operate safely without continuous human oversight. Andon Labs runs actual autonomous businesses — the Andon Market store in San Francisco, the Andon Café in Stockholm, and the Andon FM radio stations operated by AI agents. It also builds benchmarks evaluating agent behavior: Vending-Bench (running a vending business), Blueprint-Bench (spatial understanding) and Drone-Bench (autonomous drone coding), along with the Pion platform for autonomously running businesses. Andon Labs' research and tools are used by leading AI labs, including Anthropic, OpenAI and Google DeepMind.
Founders
Co-founder of Andon Labs, co-author of the Vending-Bench benchmark, researcher of autonomous AI agent safety.
Co-founder of Andon Labs, co-author of the Vending-Bench benchmark, works on evaluation and safety of frontier models.
Classification
External links
