Yesterday I taught the first class of my Business Associations course. As always, I began with the question of when an agency relationship comes into existence. I taught the same old warhorses: Gorton v. Doty and Cargill.
But in mid-class I got to wondering about “agentic” AI. So, after class, I asked Claude “what tasks do agentic AIs commonly perform” and “do any of those tasks involve the AI buying or selling things on behalf of the user”?
Here’s what Claude said (AI generated):
Yes—commerce is one of the more actively developed (and more controversial) categories of agentic AI:
Shopping agents: some AI systems can search for products, compare prices, and complete a purchase on a retailer’s site given a budget or set of preferences (e.g., early versions of this have been piloted by various shopping assistants and browser-based agents).
Financial/trading agents: there’s experimentation with AI agents executing trades or rebalancing portfolios within pre-set rules, though this remains more constrained and heavily regulated than shopping use cases.
Booking and reservations: buying tickets, booking travel, or reserving services counts as a transaction-executing task and is a common target for agentic systems.
Business procurement: agents that place recurring orders (e.g., restocking inventory) based on preset thresholds.
That said, giving an AI agent the ability to actually spend money is treated with more caution industry-wide than read-only or draft-only tasks, since it involves real financial risk, authentication, and liability questions. Many current implementations stop short of full autonomy here—the AI might find options, compare prices, or prepare an order, but leave final confirmation of a purchase to the human.
“More caution” does not mean it isn’t happening. And a lot of that looks like the sort of stuff human agents do all the time.
So the question arises: If your AI agent signed you up to a contract, would you be bound by that contract in the same way you would be if a human agent had done so?