Why AI Commerce Isn't Going Anywhere (news.alvaroduran.com)

A payments engineer’s critique argues that while LLMs can technically buy things online, “AI commerce” is unlikely to become mainstream because the social, legal and economic realities of payments defeat the promise. Drawing a parallel with RPA’s hype cycle, the piece contends that AI agents face three hard limits: they can’t safely impersonate humans (authentication/authorization), they can’t reliably infer exact user intent (authenticity = permission + correct request), and they create liability when hallucinations produce unwanted purchases. A Perplexity demo shows the UX can look seamless today, but seamlessness isn’t enough if it doesn’t materially improve the buying experience or if it transfers fraud risk and chargebacks to merchants. Technically, the author urges designing for reversibility (Human Not Present payments) rather than perfect intent inference, and flags Merchant-of-Record (MoR) responsibilities—processing, chargebacks, taxes, PCI compliance—which few are willing to accept for AI-driven buys. Implications for the AI/ML community include prioritizing robust intent verification, explainability and fraud-mitigation over flashy autonomous agents, accounting for operational costs (akin to interchange fees), and building systems that accept and manage hallucination risk. The takeaway: AI agents may assist search and delegation, but widespread, autonomous AI-driven commerce faces structural challenges that are not solved by better models alone.