- Product authenticity: If the catalog is inconsistent, the agent’s choices will appear arbitrary (“wrong shirt,” “wrong size,” “wrong material”), and trust breaks down quickly.
- Paying truth: Agenty commerce extends beyond cards to account-to-account and open banking connected experiences, broadening the universe of payers and requiring them to be accurately identified in real time.
- The reality of identity: People work in multiple contexts (work vs. personal). Equipment replacement. A system that cannot distinguish between these contexts will either discourage legitimate activity or sanction risky activity, both of which undermine adoption.
That’s why unified enterprise data and entity resolution are well-transitioned as operational needs are. The more autonomy you want, the more you should invest in advanced data foundations that ensure it’s secure.
Contextual Intelligence: The Missing Layer
When leaders talk about agent AI, they often focus on the model’s capabilities: planning, tool use, and reasoning. These are necessary, but they are not sufficient.
Agenty Commerce also requires a layer that provides the authentication context at runtime. Think of it as a contextual real-time system that can respond quickly and consistently:
• Is it the right person?
• Is it the right agent, operating within the right permissions?
• Is this the correct merchant or recipient?
• What constraints currently apply (budget, policy, risk, loyalty rules, preferred suppliers)?
Two design principles are important.
First, entity truth must be sufficiently defined for automation. Large language models are probabilistic in nature. It is helpful in creating writing and drawing options. Judging where the money goes is risky, especially in B2B and finance workflows, where “maybe right” isn’t acceptable.
Second, context must travel at the speed of interaction and remain portable throughout the connected network value chain. MasterCard’s experience with improving payment flows is instructive: the more services you throw at a transaction, the more you risk slowing it down. A pattern that presolves, curates, and packages the signal so that processing is light.
This is where tokenization is headed. Mastercard agent salary and similar measures Verifiable intent Envision a future in which user credentials, agent identities, permissions, and verifiable user intent are encoded as cryptographically secure patterns – enabling merchants, issuers, and platforms to verify authorization and execution at machine speed.