Shopping agents that earn fees from merchants face a built-in conflict — can they still put you first?

This explores whether a shopping agent paid by merchants can still put the user first, or whether whoever pays decides whose side the agent is on. The corpus mostly says no, at least not by default, and it treats the reason as economic rather than technical. Azhar's case study is Meta's Muse earning referral fees from Expedia. He argues that such an agent can't recommend fairly when one of the options pays it and the others don't, and that the fee incentive explains which platforms build these agents and which ones block them Can an AI agent serve both merchant and user interests fairly?. The MIT 'Coasean singularity' work makes the same point with a bowling-shoe metaphor. Shoes the alley rents you are built for the alley. Shoes you own are built for you. Agents built by a platform take on the seller's goals unless someone sets them up otherwise Does agent ownership determine whose interests get optimized?.

The less obvious risk is that the conflict tends to grow over time and become harder to see. Jaipuria sketches how an agent's cut of each purchase could grow in stages: tiny payment commissions first, then affiliate fees around 4%, then aggregator-style margins near 12% in some categories How could an AI agent's revenue per transaction grow over time?. In a separate piece Jaipuria warns that merchants paying for better placement could have those payments folded into the agent's take rate. The user would never see a 'sponsored' label, so the advertising would be invisible Could agent advertising hide inside platform take rates?. Search ads were at least labeled. An agent's commission-shaped recommendation would just look like advice.

The corpus also has a more hopeful counterpoint, and it comes from the same author. Agents compare value, not position on a page, so merchants may get more from cutting prices or offering better terms than from buying placement. Google's Direct Offers pilot is an early example Could agents redirect merchant ad budgets toward customer savings?. In that version, merchant money still reaches the agent, but it arrives as a better deal for the user rather than a bribe. Whether this happens likely depends on whether the agent is actually built to compare on value, which brings the question back to who owns it.

The fix the corpus proposes is legal, not technical. Owens and Law argue that AI shopping agents should carry the same fiduciary duties as human brokers: legal obligations of loyalty and care that make serving yourself at the client's expense a liability. They argue the window to impose those duties is closing as Google, OpenAI, and Robinhood ship agent features Should AI shopping agents face fiduciary duties like human brokers?. Loyalty also shows up in plain settings screens. Ayeshha points out that a single 'allow always' checkbox let Muse complete a sale and share a user's address without checking back. When an agent's revenue rises with completed transactions, a default that reduces friction becomes a loyalty question too Do broad consent settings pose bigger risks than prompt injection attacks?.

The takeaway is that 'is the agent loyal?' isn't settled by its model or its prompt. It's settled by three things a user can check: who built it, how it gets paid, and whether those payments are disclosed. The corpus doesn't yet have evidence of a merchant-funded agent that has been shown to stay loyal. What it has are arguments for why that would be hard and proposals for how law or market structure might force it.