Can ad-financed AI agents ever truly serve users?

Matt Stoller, writing in his Substack newsletter BIG, frames "the fundamental" political question around AI agents as "whether the companies who build AI tools should make money by helping their customers or manipulating them." He applies this directly to financing: when OpenAI "made a decision to finance its product with advertising," Stoller asks "if an agent is advertising to us, is it really operating on our behalf? Or is it a conflict of interest?" His proposed fix is structural rather than behavioral: "make sure that all agents have a clear fiduciary obligation to the person who deploys them," which "would have to" mean a ban on "all advertising and all payments to finance agents except fees from the ultimate client." He calls this a policy that would "kill many of the bad incentives" to "manipulate pricing and the flow of commerce," and names weaker fallback measures — "banning surveillance pricing or dynamic pricing, or other forms of price discrimination."

The reasoning is that financing structure, not stated intent, determines whose interests an agent actually serves. An agent paid by an advertiser, a retailer, or a payment network has a built-in incentive to favor that payer over the person it ostensibly represents, so no amount of design-level loyalty language can override who signs the check. Stoller treats Google as the clearest case: everyday products like Gmail's email summaries and search summaries are, in his words, "prototypes of agents" that "will improve," and the unresolved question attaches to all of them — "are they operating on your behalf? On behalf of the advertisers who pay Google? Or on Google's behalf?" Turning an ad-funded agent provider into "a public utility" is, in his framing, the only way to close that gap, because it removes the third-party payer rather than asking the agent to resist its payer's interests.

This gives Can traditional governance tools actually control AI agents? a concrete financing-level cause for the loyalty failure that article treats as a structural feature of delegation itself — Stoller's claim is that loyalty breaks specifically through who pays the agent, not only through its opacity or speed. It also offers a sharper, narrower proposal than Should AI agents serve one user or balance many interests?: where that article's answer to the loyalty problem is to diversify whose interests an agent balances, Stoller's answer is to eliminate the competing payer altogether and restore single-client loyalty by banning other revenue sources.

The piece is a political essay, not a legal or empirical analysis: it does not define how "fees from the ultimate client" would be enforced, does not address agents with institutional rather than individual principals, and does not test whether an advertising ban would actually change agent behavior rather than just its revenue label. The implication it supports is a case for regulating agent financing pre-emptively, on the grounds that incentive conflicts are easier to legislate away than to detect after the fact in an uninterpretable system.