A multiplier in search of a denominator
Thirty-eight times is the kind of number that ends arguments. It should not, because on its own it contains almost no information about scale. Thirty-eight times a hundred purchases is 3,800. Thirty-eight times a hundred thousand is a market. Stripe holds the underlying figures, but in the sources reviewed — its own 29 September blog post and its comments reported by Upstarts Media — it has published the multiplier without the absolute purchase count or dollar total, which leaves outsiders unable to establish which of those scales this is. Secondary summaries of Upstarts' report describe the growth as starting from a small base, but no absolute starting count was provided in the material reviewed. Dan Hill's strongest quantification on the record is that the volume is "becoming very material." Material to Stripe, a company that processes payments at global scale, could still be small in absolute terms. Nobody outside the company can tell.
The two sources also establish different things, and they should not be blended. Stripe's own announcement documents the Link agent-purchase infrastructure and names the participating agents: Muse, Meta's personal agent, Grok Bot and Instinct. Upstarts separately reported the September spending pattern, including the average purchase of about $100 and retail as the largest category. Together they suggest agent checkout is being used for real transactions, not just demonstrations — but the absolute counts and dollar totals remain undisclosed in the sources reviewed.
The basket arithmetic, with its condition attached
Transactions reportedly grew 38-fold and volume 20-fold. If the two multipliers cover the same comparison period, ALFW's arithmetic implies the average purchase value fell to about 53% of its earlier level — a decline of roughly 47% — and an average of about $100 now would imply an average near $190 earlier. That is arithmetic on the reported multipliers, not a figure Stripe published.
The condition matters, because the reporting does not establish that the two windows match. Stripe's own post describes its 38x as growth "over the past month" as of 29 September. The September breakdown comes from Upstarts' later report, and the comparison periods behind its multipliers sit in the part of that article behind its paywall. So the smaller-basket reading is a conditional calculation, not a confirmed trend — and even taken on its own terms, the arithmetic implies a lower average purchase value without establishing why it fell.
The controls are the actual product news
Under the growth figure, Stripe's 29 September post announced three changes that matter more to anyone actually letting an agent spend. Link now supports incremental authorization, so an agent can ask to raise an approved amount when tax or shipping pushes a total past the original estimate, instead of starting over. Agents can, with a consumer's permission, read that person's purchase history through Stripe's financial connections to make recommendations; Stripe says the data covers more than 12,000 financial institutions, covering over 97% of US bank accounts. And purchases made by agents through Link may now carry purchase protections, covering things like accidental damage and returns, with Muse the first agent to offer them.
The control that would change the trust calculation most is the one that does not exist yet. Stripe says it plans spending controls that would let a consumer set an agent a budget and authorise purchases without approving each transaction individually; the 29 September post gives no launch date. Until they arrive, the degree of oversight on any given purchase depends on the existing payment and agent setup — and the published figures do not establish that approval requirements explain the average purchase size.
What would settle it
One disclosure would do it: an absolute transaction count for a defined period, published alongside a comparable volume figure. With those, the 38x becomes a market size, the $100 average becomes a total, and "material" becomes checkable. Without them, vendor multiples in agent commerce remain hard to compare, and the honest summary is narrower than the headline: rapid growth, undisclosed absolute scale. Watch for the first absolute figure, from Stripe or from a competitor such as PayPal or Shop Pay, both of which Hill named as expected rivals. The first vendor to publish a total will set the scale everyone else has to answer to.
A note on sources
The 38x figure and the product announcements are Stripe's own, from its 29 September 2026 post, which ALFW read in full. The September breakdown — 20-fold volume growth, the $100 average and the category split — comes from Upstarts Media's 18 October report. ALFW read the original's free portion, which confirms the agent counts, the 300 million user figure, Hill's "becoming very material" comment and the Anthropic/OpenAI status; the detailed breakdown sits behind Upstarts' paywall, so those figures are reported here as carried by Techmeme and secondary summaries of the report, and were not independently verified against the original text.
Sources
- [1] Stripe, "Helping personal agents shop more intelligently and reliably with Link" (29 September 2026) — the 38x figure in Stripe's own words, the named agents, incremental authorization, financial insights, purchase protections with Muse first, and spending controls described as planned; read in full by ALFWRead source
- [2] Upstarts Media, "Data: How AI Assistants Are Spending Real Money" (Alex Konrad, 18 October 2026) — ALFW read the original's free portion, which carries the agent counts (about 70 in September, about 200 now), the 300 million Link users, Hill's "becoming very material" comment, and Anthropic and OpenAI not yet using Link for agentic commerce; the detailed September breakdown is behind its paywallRead source
- [3] CryptoGames (19 October 2026), secondary summary of the Upstarts report — the paywalled breakdown figures as carried in secondary coverage: 20x volume growth, the ~$100 average, retail at 36% of transactions and 32% of volume, travel at 28% of volume, and the small-base descriptionRead source