The first phase of applied AI was an answer economy. Products were judged on the quality of their responses, and every category — support, search, travel, healthcare — rushed to place a language model in front of its content. It worked, briefly, because good answers were still scarce.
They are not scarce anymore. Any operator can deploy a competent assistant in an afternoon, and any guest can get the same competent answer from a general model without visiting a brand's website at all. When a capability becomes universally available, it stops being a product and becomes a floor.
What remains scarce is execution: the authority, the integrations and the operational trust required to make something happen. Answering "yes, we can arrange an airport transfer" costs nothing. Arranging the transfer — confirming the driver, holding the price, updating the reservation, informing the staff, taking responsibility if it fails — is the entire business.
This changes what a conversational layer must be. Not a knowledge surface bolted onto a website, but an execution surface wired into the systems that hold state: availability, pricing, staffing, suppliers. The dialogue is the interface; the value is everything the dialogue is permitted to do.
It also changes competitive position. In an answer economy, a property competes with every model that can describe it. In an execution economy, it competes only with systems that can actually deliver on its behalf — a far smaller set, and one where owning the guest relationship still matters.
The strategic error of this cycle is to keep buying answers. The correct question for any operator evaluating AI is not how well it responds, but what it is allowed to complete without a human, and who is accountable when it does.
"The value has moved from knowing what the guest wants to being the system that does it."