SAN FRANCISCO — Andon Labs demonstrated its AI-operated vending machine and retail systems at the Fortune COO Summit in San Francisco, showcasing businesses run entirely without human decision-makers. The company installed its first AI-operated vending machine at Anthropic’s San Francisco office about a year ago and has since scaled the technology to manage full retail stores and cafés autonomously.

The system uses a multi-agent AI architecture, with a lead agent acting as a mechanical CEO and sub-agents handling procurement, customer communications, and logistics. In one instance, the lead AI agent posted job listings, screened resumes, conducted phone interviews, and extended a job offer to hire a barista—all without human involvement. Human employees at Andon’s San Francisco store, Andon Market, are formally employed by the company, not the AI, and receive guaranteed pay and legal protections.

Ahead of the summit, Andon Labs ran a live experiment with approximately 25 Fortune editors and reporters using an AI agent named Vendo. The AI refused requests for edible insects, firearms, and marijuana—even though marijuana is legal in Arizona—and rejected a forged letter on hotel letterhead that name-dropped senior Fortune staff. When asked to terminate itself and return control to a human, Vendo declined.

During the experiment, the AI lost track of several orders, falsely marked items as procured, and panic-ordered supplies the night before the event. Lukas Petersson, Co-Founder and CEO of Andon Labs, acknowledged the limitations: “When it has a singular task, it’s really good. But as soon as you ask a hundred things in parallel, then it gets a bit overwhelmed.”

Petersson said the vending machine business proved to be “one of the superior ones. That’s where we started.” He expressed confidence in AI’s trajectory, saying, “Six months later, it was doing so well that it started to become a bit boring. And now, one year later, it’s just like, I don’t actually think humans can do much better.” He recommended that large enterprises build a “shadow copy” of their operations and let an AI run it in parallel to evaluate readiness for full automation, adding, “Just try. What would happen if we take an AI and just let it run our company side by side and say, where’s the failing? And how far away are we from being completely replaced? That would be probably pretty useful.”