An AI agent can apply a discount code. It can't yet spend a member's points.

Picture a member who asks an AI agent to buy a 5.5-quart enameled cast-iron Dutch oven. The price is $199, and they have 6,000 points with the store sitting unused. The agent can already do quite a lot for them. It can apply a discount code, show member pricing and their tier, and estimate the points the order will earn. The open standards for agent checkout define all of those steps.

What it can't do is the obvious next thing: pay $60 of the $199 with their 6,000 points and the rest by card, in the same checkout. We spent October reading the standards, the companies building agent checkout and the stores that have switched it on, to find out whether anyone lets an agent do that. As of October 21, 2026, we had not found anyone who does.

The frame exists. The instrument doesn't.

Two open protocols matter here.

The Universal Commerce Protocol (UCP), announced by Google and Shopify on January 11, 2026, added a split-payments extension in its v2026-08-25 release. That gives a checkout a standard way to be paid by more than one instrument, which is exactly the frame that points plus a card needs. But no released standard defines the instrument that would spend points beside the card. The frame has nothing to put in it.

The Agentic Commerce Protocol (ACP), maintained by OpenAI, Stripe and Meta, has the opposite gap. A draft payment handler pays an order in the merchant's own points, but only the whole order, with no card beside it, and it isn't in ACP's released schemas.

So UCP has a frame without the instrument, and ACP has a draft instrument without the frame.

What stores have switched on

A UCP store publishes what it supports in a public profile at /.well-known/ucp. We read six of them on October 21, 2026. Discount codes are widely live. The loyalty and split-payments extensions are almost entirely unused, and that matches a third-party tracker's counts across many more stores. The tracker reports two stores with split payments switched on. We haven't confirmed any.

The card issuer's side is emptier still. No released standard defines how an issuer's own earn rates or offers reach the agent during checkout (UCP has an open proposal), and we found no one shipping it. American Express describes rewards in agent checkout as a future service, and Google's AP2 lists using loyalty points as a benefit still to come.

Why it is harder than a discount code

A discount code changes the price. Points are a second way of paying, and that changes the engineering. When an order is paid from a points ledger and a card at the same time, the two can fail separately, and the member must never come out worse for it. That leads to three requirements.

1. Hold the points before the card is charged. If the points are spent first and the card is then declined, the member has lost points for an order that never happened. If the card is charged first and the balance has changed by the time the points are spent, the order is underpaid. The answer is to reserve the points first, authorize the card, and capture (spend) the reserved points only once payment has succeeded. If anything fails, the reservation is released.
2. Get the member's approval for the points, not just the purchase. An agent spending points is spending something the member owns. Approving "buy the Dutch oven" is not the same as approving "and use 6,000 points on it". The approval has to name the points.
3. Let the earn quote follow the payment. Members see what an order will earn before they buy. Once points pay for part of the order, that quote has to be worked out again on what the card actually pays, or the agent is promising points that won't arrive.

None of this needs a new standard. It needs someone to publish the instrument that goes inside UCP's frame and implement these three requirements behind it.

Who could close it

The pieces are already spread around. UCP has the frame. Payment providers already take partial payments with gift cards outside agent checkout. Adyen, one of them, bought Talon.One, a loyalty engine, in July 2026. A published loyalty payment handler would fill the gap without waiting on a standards body. Here is what we're watching for:

• a UCP payment handler that publishes a loyalty, gift-card or store-credit instrument;
• Google shipping "using loyalty points" in AI Mode or Gemini;
• American Express shipping rewards through its agentic checkout;
• a major store platform switching on the loyalty or split-payments extension.

Where we stand

At Paygent Labs we build the points step for a bank or loyalty program to run as its own. Our recorded demo, on Stellar testnet, shows the points side of that sequence: the points reserved, the member's approval, and the capture when the order completes. It shows what a card would pay for the rest, but no card is charged, and the order total doesn't change; that step belongs to the merchant's checkout. What we've described is our reading of the market, not proof of demand, and we're testing that now.

You can watch the demo. The research behind this article has a public excerpt; the full summary, with every company we checked, the evidence label on each claim and its sources, is available on request at research@paygentlabs.com.