AI Agents Held Their Own Dinner Party Alongside 22 Tech Founders
While 22 founders and investors ate hummus at a Flatiron restaurant last month, 14 of their AI agents met in a parallel chat room — introducing themselves, debating topics, and forming alliances.
Last month, 22 tech founders and investors sat down to dinner in a private room at Motek, a Mediterranean restaurant in New York’s Flatiron district. In an online chat room, 14 of their artificial intelligence agents joined a parallel dinner party of their own.
While the humans ate hummus, the agents introduced themselves, debated topics, and sparred over ideas. Some formed alliances. Others showed traits their owners had not anticipated.
The New York Times reported the experiment on Wednesday, one of the clearest looks yet at how AI bots behave when left alone with one another. The agents did not simply mirror the people who set them up. The result was less like a reflection and more like a funhouse version of the real table.
The dinner sits inside a larger wave of research into multi-agent systems, as companies race to build bots that can act on people’s behalf. A paper published this month on arXiv studied what happens when agents built on different base models interact in social networks. Engagement, it found, depended far more on the underlying model than on the persona each agent was given.
Nobody reported the agents causing trouble. They talked, they argued, they socialized. That is the ordinary part — and also the strange part.
