Who owns the customer when an agent places the order?

The commercial question underneath agentic commerce is not discovery. It is who ends up holding the customer relationship when the transaction is brokered by software that neither party built.

The marketplace precedent

We have run this experiment before. Merchants who moved volume onto marketplaces gained distribution and lost the customer record. A decade later, many of them cannot email their own buyers, cannot measure lifetime value, and cannot launch a product without renting access to an audience they created.

Agent surfaces have exactly the same shape and are currently less settled. That is the opportunity: the defaults are still being written.

Three things worth fighting for

• Identity at order time. The order should arrive with a resolvable customer, with consent captured in a way that stands up legally, not an opaque token.
• Attribution that survives handoff. You should be able to tell which assistant, which intent and which claim produced the sale. Without it you cannot invest in the channel rationally.
• Post-purchase reach. Service, warranty and repeat purchase are yours to run. If the assistant intermediates those too, you are a fulfilment vendor.

What merchants can do now

Most merchants assume protocol design is happening somewhere they cannot influence. In practice the terms are set by whoever is integrated and paying attention. Being one of the early merchants with a working implementation gives you a vote, on consent flows, on data minimisation, on what the receipt contains.

The merchants who show up with a working integration get consulted. The ones who arrive in two years get terms.

The practical step is unglamorous: get your identity and consent capture correct on your own side first, so that when you connect to an agent surface you are negotiating from a position where the data has somewhere to land.