Oct 15, 2026 · 51 min
AI Agents Move From Recommendations to Real Purchases
When AI Agents Start Spending Your Money | Alex Rampell & Terry Angelos
Once agents can spend on a person’s behalf, payment security, delegated authority, merchant control, and liability become central consumer questions.
- 1Credential vaults and tokenization could let agents transact without repeatedly exposing card details to merchants.
- 2Agentic tokens may preserve network protections, but identity, intent, mistakes, and liability remain unresolved.
- 3Retailers must decide whether agents expand discovery and sales or weaken control over scarcity, advertising, and customer relationships.
Don't miss
Rampell’s breakdown of identity, authorization, and intent shows why an agent’s incorrect purchase creates a fundamentally difficult liability question.
The brief
AI agents are moving beyond recommending products toward completing transactions, raising a basic infrastructure question: how can software spend without exposing credentials or exceeding a user’s authority?
Terry Angelos explains how VGS vaults and tokenizes payment credentials for merchants, neobanks, and frontier AI labs, while supporting updates and routing across processors.
The conversation reframes agent wallets as more than saved cards: they must connect consumers, agents, merchants, and payment networks while preserving trust and dispute protections.
Rampell separates identity, authorization, and intent, exposing the unresolved liability problem when an agent hallucinates, misunderstands a request, or follows rules imperfectly.
Agentic commerce could help with difficult purchases and reshape discovery, subscriptions, scarcity, and advertising—but retailers may resist if agents weaken their customer relationships.