Agents

When AI Shops For You, Taste Becomes the Product

Mastercard transactions processed in 2025.

Every one of them is about to change.

The agent layer

Mastercard launched Agent Pay: AI agents that initiate and complete purchases on your behalf. Not a chatbot. Not a recommendation engine. A buyer. Software that holds your card details, understands your preferences, and executes transactions without you lifting a finger. The infrastructure for autonomous commerce is no longer theoretical. It shipped.

Mastercard's Q1 2026 restructuring charge. Roughly 4% of the global workforce, reallocated toward AI, services, and crypto.

When a company restructures around a thesis, it is no longer a bet. It is a commitment.

Trust becomes infrastructure

Mastercard is spending billions becoming the verification layer for agentic commerce. Identity confirmation, fraud scoring, biometric authentication. The agent needs to be trusted before it can act. This is not a feature release. It is the construction of an entirely new trust architecture, one where machines vouch for machines on behalf of people.

BVNK acquisition. Connecting on-chain rails to fiat payments.

Building the plumbing for agents to move money anywhere.

Commerce without browsing

When an agent buys for you, the storefront becomes irrelevant. No homepage. No product grid. No add-to-cart. The entire visual layer of e-commerce was built for human attention. Agents do not have attention. They have instructions. The question is no longer "how do we get them to the checkout?" It is "how do we get chosen by the algorithm that never visits our site?"

Taste as the input layer

The agent needs instructions. Not just "buy milk" but "buy the oat milk from the brand I actually like, in the size that fits my fridge shelf, from the shop that is not out of my way." That specificity is taste. And most people cannot articulate it. The gap between what you want and what you can express to a machine is where the next generation of products will live. I have been building for exactly this gap. Curio maps cultural taste outward. Trove maps it inward. Taste OS turns it into a system. These are not side projects. They are prototypes for the input layer that agentic commerce will need.

The curation premium

Mastercard's Virtual C-Suite gives small businesses executive-level intelligence from transaction data. The pattern is clear: data becomes insight becomes action. The brands that survive agentic commerce will be the ones whose taste profile is so well-defined that an agent can represent them faithfully. Not the cheapest. Not the fastest. The most legible.

The model behind the model

Mastercard is building a proprietary Large Tabular Model, a generative AI engine trained not on text or images but on structured financial data. Hundreds of billions of anonymised transactions. Merchant location data. Fraud patterns. Loyalty behaviour. This is the intelligence layer that will power agentic commerce. And the brands whose taste signals are legible to this model will be the ones the agents choose.

Companies in Mastercard's Crypto Partner Programme.

Building the rails for agent-native payments.

Culture is the last moat

Price can be compared instantly. Availability can be checked in milliseconds. Delivery can be optimised algorithmically. The only thing an agent cannot replicate is the cultural meaning of a choice. Why this, not that. Why now, not later. Why here, not there. Cultural context is the one input that resists automation. It is messy, subjective, and constantly shifting. Which is exactly what makes it valuable.

Who briefs the agent?

This is the new strategic question. Someone will define the taste parameters that guide agentic purchasing. It could be the consumer, but that requires too much effort from people who cannot articulate what they want. It could be the platform, but that creates a dangerous conflict of interest. Or it could be the brand. And that is the opportunity. The company that teaches the agent what good looks like wins the transaction.

The brands that survive will be legible to machines and desirable to humans. Most will only manage one.

Agents handle the transaction. But what happens when the transaction disappears entirely? The Invisible Layer argues the best experiences remove themselves. Together these decks describe a world where the brand is felt but never seen.

Build the taste layer

The companies investing in cultural intelligence, signal detection, and taste profiling today are building the input layer for tomorrow's agentic commerce. This is not futurism. Mastercard shipped Agent Pay to every US cardholder. Santander completed Europe's first live AI agent payment. It is already here. The only question left is whether your brand is legible to the software that is about to do the buying.

Thank you.

mikelitman.me · hello@mikelitman.me