Stripe’s own blog says agentic purchases made with Link rose 38x over a month. Its president told a podcast that agentic commerce “started going vertical” in September. Four published accounts handle those two statements differently, and the gap between them shows which numbers a finance team should ask for.
What Stripe published and what its president said
On 29 September, Stripe posted a product update for Link’s wallet for agents. It names Muse, Meta’s new personal agent, Grok Bot and Instinct as agents that use the wallet, and states that “over the past month, agentic purchases made with Link increased 38x.” The post gives no purchase count, dollar value or merchant total behind the multiplier.
Two weeks later, John Collison, Stripe’s president, talked agentic commerce on the Cheeky Pint podcast he hosts, in a conversation with Meta CEO Mark Zuckerberg and Meta chief AI officer Alexandr Wang. Payments Dive quotes him saying agentic commerce “basically started going vertical” as of September. Collison also said that “computer use was the last big unlock that we needed to make it happen.”
Advertisement
Four accounts, four emphases
Payments Dive, in a piece by Lynne Marek, reports the remark as a company disclosure and says “Collison didn’t provide any specific figures”. The piece places the comment inside the larger conversation about Meta’s Muse agent and Stripe’s $7.5 billion purchase of OpenRouter in August, a deal we covered in Stripe Turns AI Token Routing Into Payments Infrastructure.
Forkast built its story around the missing numbers. Its headline says “He Has Not Shown the Numbers,” and its text separates what the infrastructure shows from what the claim shows. It also cites two survey figures for context: Product.ai data from April 2026 putting consumer trust in AI executing purchases without manual verification at 14%, and PYMNTS Intelligence data from September in which 93% of merchants say AI or agent providers should bear the loss when an agent makes an incorrect purchase. We have not seen either survey and repeat them as Forkast reports them.
Bloomberg’s account of the Amazon block on Muse, carried by The Star, takes the opposite view of the same market. It says “consumers are mostly using bots to research products rather than let them make purchases,” which reads against Wang’s description, quoted by Payments Dive, of Muse as a service where “people are fully delegating purchases to the agents.”
VKTR covers the Amazon side in more detail. It reports that Amazon started blocking Muse on 20 September, that Amazon cites its Conditions of Use, and that Amazon says it asked Meta to remove Amazon from the Muse experience. It also reports Meta’s earlier statement that Muse has no visibility into users’ passwords or payment methods.
Where the accounts disagree
The accounts split on three points.
The first is whether a figure exists. Payments Dive and Forkast both say Collison gave none. Stripe’s Link post, published two weeks before the podcast, contains one: the 38x multiplier. Neither article mentions it. A multiplier is a rate of change and Collison’s remark was a statement about volume, so the two do not answer the same question, but a reader of either article would conclude Stripe has published nothing numeric.
The second is the size of the market. Wang’s “fully delegating purchases” description and Bloomberg’s “mostly using bots to research” sit three weeks apart and describe the same product category. Both can be true if delegated purchases are a small share of agent activity. None of the four accounts gives the share.
The third is where the constraint sits. Forkast puts it on disclosure and consumer trust. Bloomberg and VKTR put it on merchant access, with Amazon saying in the words of spokesperson Lara Hendrickson that “third-party applications that offer to make purchases on behalf of customers from other businesses should operate openly and respect service provider decisions about whether or not to participate.” Zuckerberg, quoted by Payments Dive, offers the counter-signal that “way more businesses are calling us asking to integrate deeply, and very few really who have tried to block it.”
What the coverage adds up to
In our reading, the four accounts describe three separate layers that share one headline. The first layer is demand, where Stripe’s 38x and Collison’s “vertical” apply. The second is access, where Amazon’s block and Zuckerberg’s integration calls apply. The third is liability, where Link’s new purchase protection and the 93% merchant survey figure apply. Each account reports one layer and treats it as evidence about the others.
The layers also have different owners of the data. Stripe holds the demand data and has released a ratio. Merchants hold the access decisions and have released statements. Nobody has published loss rates, which is the number a bank or acquirer needs to price chargeback liability. Stripe’s post says qualifying Link transactions are covered for “accidental damage, lost items, price drops after they buy, no-fee returns, and return guarantees.” It also plans spending controls that let a consumer set an agent’s budget “without having to approve each individual transaction”. Dropping the per-purchase approval moves the approval step from the consumer to a rule, which changes who is liable when the rule is wrong. We made the case for a refund standard in Agent Shopping Needs a Refund Rule, Not a Growth Rate, and the budget feature makes that question more pressing.
A 38x multiplier over one month fits a small starting base as well as a large one. Without the base, it supports the claim that something changed in September and nothing about how much money moved. The same limit applies to every vendor release that reports growth as a multiple.
The Link post adds one detail that matters to banks. It says Link gives agents consumer-permissioned access to transaction data from more than 12,000 financial institutions covering over 97% of US bank accounts, through Stripe’s Financial Connections open banking product. An agent that sets a budget from spending history is reading the same account data a bank holds, and the consumer decides which accounts to share and can revoke access at any time. VKTR adds that Amazon and Meta are also business partners, with Amazon products purchasable inside Facebook and Instagram since 2023, so the access dispute sits inside an existing commercial relationship.
What it means for the finance leader
A bank, acquirer or issuer deciding how much to invest in agent payments needs three figures that the coverage does not supply: the dollar value of agent-initiated purchases, the share that settle without a human approval step, and the dispute rate on those purchases. Ask processors for those three when a vendor presents a growth multiple.
Treat merchant access as a separate risk. Amazon’s block shows that an agent can have a working payment credential and still be refused at the storefront. Product plans that assume universal merchant acceptance should be tested against the merchants that matter to the customer base, and against the contract terms those merchants publish. Our earlier look at consumer attitudes, Consumers Will Let AI Shop. Not Pay Yet., gives the demand-side baseline to compare against.
Finally, set a review date. Stripe says spending controls are planned, and they would change how much a consumer delegates. If they ship, the first published volume or loss figures will say more than any growth multiple.
Source: Stripe, “Helping personal agents shop more intelligently and reliably with Link”
