Owens describes an experiment in which AI shopping agents were sent to find the cheapest flights, health plans and graduate programs. Her broader argument on the episode is that agents calling themselves your personal assistant should be legally bound to work for you, not for whoever pays for placement.
…you log on from the Bay Area, where they think you're more affluent and a lot more, right? So, this is going to happen. We have now a really interesting study of Agentic Commerce that just came out. They did an experiment with 325,000 agents, and they asked the agents to go find the cheapest item.
Okay.
Even when directed explicitly to go get the cheapest item, the agents rooted around in your email inbox, identified that you could indeed afford more than the cheapest option, and brought back a more expensive option.
So, an agent that's working on my behalf, are you saying would not actually be working on my behalf?
That's exactly right. That's what the experiment discovered. They asked the agents to buy flights. These were pretty big ticket items, health insurance plans, and to select a graduate program. And even if you said, bring me the cheapest health insurance plan, once the agent figured out a little bit…
What's your thesis for what's going on there? Because I can imagine maybe something like Muse. I mean, Meta is also an advertising platform, right? So, maybe it wants to route some business to particular companies that advertise a lot on its platform, but maybe there's something else going on too.