Checkout does not disappear — it moves

Part 2 of four on agentic commerce. The series opens with Stripe says the checkout page is dying.

The easiest way to misread “the checkout page disappears” is to think checkout disappears. It does not. Everything it does still has to happen — nobody can see it any more, that is all.

What a checkout page actually is

We think of checkout as a form. It is really a set of decisions that happen to have an interface:

• is this item genuinely in stock, right now
• what does it cost for this customer, with this discount, in this country
• which VAT applies, and is the customer registered
• can it be delivered there, and when
• is this customer allowed to buy on invoice, and for how much
• what happens if payment is declined halfway through

The form is the least interesting part. It is only where the decisions get shown to a human.

What happens when the page is removed

The decisions remain. They simply have to be made without anyone able to look at them first — and answered fast enough that an agent does not give up and move on.

Which means things that could be handled sloppily behind an interface no longer can:

Stock has to be true, not approximately true. A human who sees “1 left” and hits an error in checkout gets annoyed. An agent that hits an error switches supplier and does not come back.

The price has to be complete in the answer. Not a base price plus a discount calculated at a later step. The agent compares on the number it is given, and if your number is a list price that nobody pays, you lose on a price you never intended to charge.

Shipping cost has to be obtainable before the purchase. This is the most common gap. Plenty of stores cannot say what shipping costs until an address is entered at step three. An agent asks before.

Error messages have to mean something. “Something went wrong” is an acceptable answer to a human who can try again. It is not an acceptable answer to a machine deciding whether to wait, switch variant or abandon.

Where the logic ends up

In one place, is the short answer. That is the real change.

In many stores today the checkout logic is scattered: VAT in one extension, shipping rules in another, customer discounts in the ERP, and a special rule for one particular customer sitting in the theme because that was quicker at the time. It works as long as there is a page where it all meets and renders.

Without that page there is no meeting point. The logic has to live somewhere askable — in the platform, reachable through the same API that answers the agent. Fundamentally this is not an agentic-commerce question. It is a question of whether your store has one truth or five.

Which is why we think this deserves attention despite the uncertain timeline: consolidating the logic pays off whether agents arrive in 2027 or 2032. It makes your current checkout faster and your bugs fewer in the meantime.

The next part is about the segment where this is already reality, and has been for twenty years.