The free agent and the missing reader

In mid-September a payments analyst pulled me into a thread about Meta's Muse, the personal agent that is supposed to shop for you, cancel the subscriptions you forgot about and, eventually, run a small business on your behalf. The comments were the best part. Everyone was arguing about the wallet: which wallet Muse pays through, whether that wallet is strong or weak, and the closing question, what if Meta simply buys PayPal?

I read the whole thread and had the same reaction I have had for twenty years whenever a payments product ships with a short list of partners it can pay through. I don't think it will work the way they think, and I'll explain why from the merchant's side, because that is the side I know.

What building one API into six hundred processors taught me

I spent a good part of my career building orchestration, which is the polite word for one integration that lets a merchant reach many processors without rewriting their checkout. By the time I was done it was one API into more than six hundred of them. People assume you build something like that because you love integrations. Trust me, nobody does.

You build it because merchants will not switch processors to adopt your product. The conversation always went the same way. The merchant liked the feature, agreed it would make them money, and then asked which processor it required. If the answer was "not yours," we were done. And they were right to walk. Switching means the old acquirer keeps holding your reserve through the chargeback tail, often six months after you leave, while the new one underwrites you from scratch and may set a reserve of its own. Two companies holding your money at the same time, plus a re-certification, plus a new month-end close. Merchants understand that cost better than the vendors selling to them do.

So when I read that Muse pays through a short list of partners (Stripe Link, Shopify's checkout, a couple of bots), I did not think about which wallet is weak. I thought: any seller outside that list is invisible to the agent. The consumer will not notice. The consumer will see an agent that found a good price at a store that happened to be on the list, and never learned about the better price at the store that was not.

Thirty-six million is still one company's list

The figures in the Payments Dive piece that quoted the analyst: Stripe Link has about three hundred million users and roughly a million merchants where it works as a native checkout. PayPal has something like thirty-six million merchants. The analyst read that as Link being the weakest player. I read it differently. If Meta bought PayPal tomorrow, the agent would still be shopping inside one company's merchant list. A longer list, but still a list, and the best deal for you would still be at merchant number thirty-six million and one.

Someone else in the thread made the point that Meta did not pick a closed checkout by accident. They picked it so the record of what you bought stays with a partner they control. That works for Meta. It does not work for you. An agent that truly works for you has to be able to buy from anyone, through whatever processor that seller already uses. I agree with that completely, and if you agree with it too, you have just inherited a problem the wallet people never have to deal with.

The free agent's two questions

An agent that can buy anywhere needs two answers, and today nobody can give them to it.

Before the purchase: is this seller real, and is it in good standing? Not "does it have a website." Is somebody with their own money at stake willing to say this business is fine?

After the purchase: did the sale stick? Was it refunded, or disputed by the cardholder? Did the money get paid out to the seller and stay paid out, or was it pulled back by a chargeback that can arrive up to four months later?

The frustrating part is that both answers already exist. The company that takes payments for a seller, what the industry calls the acquirer, knows both. When an acquirer sees exposure in a new business, say goods that ship later or a category with a history of disputes, it delays payouts or holds back a slice of the money for a few months. The terms it sets today are a verdict from a party with its own cash at risk. After a sale, the same company knows whether a refund or a dispute landed, because it is the one on the hook if the seller cannot cover it.

Those answers sit in the acquirer's risk and settlement files. The files were built for the acquirer's own protection, and nobody outside was ever meant to read them.

The missing reader

Another commenter put it better than I had: the gap is not a missing check, it is a missing reader. The acquirer produces the signal every day; the buyer's agent has no way to ask for it.

I would add two things, from having sat on the acquiring side.

One. A seller's standing is not a fixed fact. Terms get tightened when volume spikes and loosened when a clean quarter goes by, so whatever the agent reads has to carry a date and come from the party that is still carrying the risk. A certificate a seller downloads once and shows around is worthless within days. I have watched that kind of thing get gamed.

Two. The reading has to go in both directions. A small seller is just as exposed to a buyer who refunds everything and disputes the rest as the buyer is to a seller who never ships. Airbnb worked this out years ago with a history on both sides and the right of either side to say no. When an agent pays on rails that cannot be disputed, that two-way history is most of the protection anyone is going to get, and the rest is money the seller can see but cannot touch until the goods arrive.

Why this is where the money gets decided

Meta says Muse will take "a very small cut of the value that we create." A cut of value only works if you can prove the value, and proving it means answering the second question after the dispute window closes. That answer lives with the seller's acquirer, not with Meta, and not with the partner Meta contracts with either, unless that partner happens to be the acquirer too.

So there is a real tension here. The open agent consumers want spreads the settlement record across thousands of acquirers. The closed agent Meta built keeps it with one partner Meta can sign a data agreement with. Meta solved its problem by keeping everyone inside the list, and the consumer is the one paying for the fix.

The honest way to solve it is a reader: a way for a buyer's agent to ask any seller's acquirer "is this business in good standing today," and a way for whoever charges a fee to ask "did this sale settle and stay settled," across acquirers, with a date on every answer.

Acquirers already share one file across companies. It is a negative list, the businesses that got terminated, and only other acquirers can read it. What is missing is the positive answer, with a date on it, readable from the buyer's side. The query is the easy part. The hard part is convincing the party that holds the file to answer questions from strangers, when network rules and the merchant agreement both tell it not to.

There is an incentive, though. An acquirer that publishes standing steers agent traffic toward the sellers it already trusts, which is the cheapest risk control it will ever buy. And a platform that can read settlement across acquirers gets a take rate it can invoice rather than argue about.

If you are a merchant

Ask one question about any agent that shows up at your door: whose merchants can it buy from? If the answer is "ours," you are not getting a new customer, you are paying to join their club. Don't they call that a membership?

If you run one of these platforms

Treat the reader as a product, not a compliance exercise. Whoever ships it first sets the terms.

I would like to hear from acquirers on this. You are sitting on the answer. What would it take to publish it? Happy to get on a short call with anyone working on it.

Originally published in Inside Payments on LinkedIn.