Does AI agent resistance to marketing prove buyer rationality?

In a Substack essay on behavioral pricing and AI shopping agents, Utpal Dholakia revisits a claim that predates the current wave of agents. Back in 2017, Gal and Elkin-Koren argued that "algorithmic consumers" could strengthen the buyer's side against seller persuasion by "searching widely and resisting persuasive manipulation." Dholakia notes they "also perceptively cautioned that much would depend on who controlled the algorithms running the algorithmic consumers" — a caveat he says today's "consultant and industry discussion on this issue has adopted the authors' optimistic take without acknowledging their caution."

Dholakia's own argument builds on evidence that AI shopping agents do not respond to human-targeted marketing tactics the way humans do (see Do human shopping tactics work the same way on AI agents?), but he draws a sharper conclusion from it than the studies themselves do: inconsistency is not rationality. He writes that this raises "two important follow-on questions... (1) What do machine buyers, a.k.a. AI agents, respond to instead, and (2) Who decides the machine's decision calculus?" The second question is the one that revives Gal and Elkin-Koren's caution: if the shopping agent is supplied by a browser, a platform, or an app that the seller also pays to advertise on, the fact that the agent resists human-style persuasion says nothing by itself about whose interest it actually optimizes for.

This argument sits close to Can EU consumer law protect AI agents as algorithmic buyers?, which argues from a legal-design angle that existing consumer-protection frameworks assume a human decision-maker; Dholakia's point is the economic mirror of that argument — even where the law or the platform treats the agent as the consumer's proxy, its actual loyalty depends on who built and operates it, not on whether it ignores charm prices or partitioned pricing. It also complicates Will brands succeed by building their own AI shopping agents?: if agents are controlled upstream by platforms with their own commercial interests, becoming the agent of choice may mean satisfying the platform's ranking logic rather than the end consumer's.

The excerpt does not name which actors currently control dominant shopping agents, nor does it present evidence of actual misalignment between an agent and its nominal principal — Dholakia poses the control question but does not answer it here, and Gal and Elkin-Koren's original caution is nearly a decade old, predating today's specific agent architectures and the commerce protocols the essay mentions (OpenAI's Instant Checkout, Google's Universal Commerce Protocol). The implication the excerpt supports is narrower than agents are compromised: findings that agents resist human-style manipulation cannot, by themselves, be read as evidence that agents serve the buyer's interest, because who controls the agent — not how it responds to pricing cues — is what determines whose interest it serves.