Instinct's AI agent is sending users unrequested shopping lists built from their past chats

Memory is what makes a personal agent worth using, and it is also what lets the agent start a sale. Kirk Plangger, chair and head of marketing at King's Business School, said what changes most is that an agent can take what it has learned about a user and start the shopping itself ^[20].

Whether that becomes a conflict depends on how the company is paid. On the Invest Like The Best podcast, chief executive Noah Shinn described a model where users pay nothing and Instinct earns a cut from merchants when it sends them a customer and helps the purchase go through ^[16]. Plangger said that even if products are chosen in the consumer's interest, Instinct could have an incentive to create more chances to shop if it benefits whenever a purchase is made ^[21]. Davood Wadi, who works as an AI research scientist at General Motors and is a research affiliate of the Desautels Faculty of Management at McGill University, put the test in one line. If Instinct makes money when a user buys what it suggests, "then yes, this is monetization, even without any ads," he said ^[19].

Plangger also urged caution. There is "no evidence that these particular recommendations were paid placements or that Instinct made money from them," he said ^[17]. Bhatia said he did not believe the picks were ads, though he called them "a bit of a turnoff" ^[11]. The evidence so far is three users who spoke to Business Insider, among dozens posting on X, and Instinct did not respond to a request for comment ^[1][9]. Plangger's caution is right about these messages, but users are judging them without the evidence either way. "I just shrugged it off as they're probably experimenting on how to make money," Tihtina said ^[15].

Shinn's own red line is drawn around what users want. "I think it would be a very dangerous world if Instinct were influencing the user's behavior to purchase something that they don't want to purchase or to subscribe to something that they don't want," he said ^[8]. He has also criticised platforms that use paid ads to persuade people to buy products they "may or may not want" ^[18]. The complaints test something else, whether the user asked, and a pick can fail that test even when the product is welcome. "I didn't ask for anything," Gandhi told Business Insider ^[2]. Bhatia said Instinct should have sought his permission before making occasional recommendations drawn from his personal history ^[12].

For a team deploying an agent this quarter, the trade-off is between reach and consent. An opt-in for proactive suggestions, the step Bhatia asked for, would reach fewer users. Unprompted picks reach all of them and leave each one to work out why a product appeared. "I have no idea why it chose what to show me," Tihtina said ^[14]. The timing of the rules matters as much as their content. I'd expect memory and disclosure rules written after merchant revenue arrives to be read against every suggestion sent before them. Instinct faces that choice while it expands its shopping tools through a Shopify partnership and rolls out Instinct Selections, a feature Shinn says draws on picks curated by chefs, designers, architects and local guides ^[4].