Agentic commerce: what to actually do, and what to leave alone

Part 4 of four on agentic commerce. The series opens with Stripe says the checkout page is dying.

The series has argued two things that look contradictory: the checkout page is not dying yet, and you should still do something. Here is the list. It is shorter than lists in this genre usually are.

Do this

1. Check that your product data can be answered with. Colour, size and material as fields, not as sentences in a description. It is the same work that makes your own filtering useful, so it pays off whatever happens with agents.

2. Find out how old your stock figure is. Not what it should be — what it is. Plenty of people discover the answer is “from last night” and that nobody decided that, it just ended up that way.

3. See whether shipping cost is obtainable before checkout. If the answer is no, it is the most concrete gap on this list, and it is already showing up in your conversion rate.

4. Count how many places a price can be decided. If it is more than one, write down which. It need not be solved this year, but it should be known.

Do not do this

Do not rebuild checkout. Nothing in this development requires it, and the checkout you have serves the customers you actually have.

Do not buy an agent extension. The standards are not settled. What is on sale today is a bet on which protocol wins, and an extension that bets wrong is an extension you get to tear out.

Do not write an AI strategy. The four points above are data quality work. They gain nothing from being called something else.

What actually decides whether this becomes your advantage

None of the points above is about agents. They are about whether your store can answer a question truthfully without a human interpreting the answer. That requirement has been there all along; the interface has been hiding it.

Which is why we think Gaybrick is right on substance and wrong on pace. The checkout page is not disappearing this year or next. But the direction is right, and the preparation happens to be the same work that makes your current store faster, more accurate and cheaper to run.

That is an unusually comfortable kind of future-proofing: it pays for itself even if the future does not turn up.