Agentic commerce finally reported its own numbers

Almost every agentic commerce figure I have written about this year came from somebody who sells forecasts for a living. Last week this brief covered two of them that were a hundred times apart, which tells you roughly what a forecast in this category is worth.

On 5 August, two companies that have to file reported instead. Shopify and Circle both put agentic numbers on the record, on the same day, under the mild legal discomfort that attaches to saying things on an earnings call [2] [4].

The figures are smaller than the forecasts and considerably more useful. They are also, every one of them, a growth rate or a market share. Not one is a level, and that omission is the story.

What Shopify said

The quarter itself was strong on any reading. Revenue of $3.6bn, up 34 per cent. Gross merchandise volume of $116bn, up 32 per cent. Free cash flow margin of 18 per cent. Shopify's own headline on the release claims 30 per cent plus growth across GMV, revenue, gross profit and free cash flow [1] [3].

On the call, president Harley Finkelstein said AI-driven traffic and orders to Shopify stores tripled year over year, and that new buyer orders arriving through AI channels come in at nearly twice the rate of other channels [2] [3]. He framed agentic commerce as additive rather than cannibalistic, which is the correct thing to say and also the thing you would say either way.

The tripling is the headline and it is the weaker of the two numbers. Things triple when they start small. I have shipped features that tripled their usage in a quarter and the honest internal description of that was "eleven people now instead of four".

The second figure is the one to keep. Nearly twice the rate is a statement about mix, not about volume, and mix statements survive scale in a way growth rates do not. It says the people arriving through an agent are disproportionately new. Whatever the absolute number turns out to be, that channel is doing acquisition rather than cannibalising a checkout that was going to happen anyway.

What Circle said

Circle's call landed the same day. Jeremy Allaire said the company "launched the Agent Stack to put programmable money at the center of the agentic economy", and reported that the Agent Stack now supports over 900 paid services, with USDC accounting for 99.3 per cent of payment volume running over the x402 agent-payment protocol [4].

Circle positioned agent activity as infrastructure utilisation rather than near-term revenue, which is a rare piece of restraint in a quarter where it would have been easy to do otherwise [4].

The 99.3 per cent is built to read as dominance. It is not quite that. A payment market in which one instrument carries 99.3 per cent of the volume is not a market anybody has competed in yet. It is a market with one shop open. That share measures how early this is, and it can only fall from here.

One date worth putting in the diary

Comments close on 21 August on the joint proposed rule setting customer identification programme requirements for permitted payment stablecoin issuers, issued by FinCEN with the OCC, the Federal Reserve Board, the FDIC and the NCUA under the GENIUS Act [5]. It was published in the Federal Register on 22 June with a sixty day window [6].

If any part of your roadmap has an agent paying in a regulated stablecoin, that consultation is the last cheap moment to say something about how identity works at the issuer level. Fifteen days.

Read from the rails

I have sat in the room where the new-thing metric gets chosen, and it is never chosen before the quarter. It is chosen afterwards, from a shortlist, by people looking for the framing that is both true and flattering. That is not a scandal. It is just what the job is.

The useful exercise on a day like this is noticing which numbers were not given. Nobody disclosed agentic GMV as a share of the total. Nobody gave a dollar figure for agent-originated volume. Both companies had that number in front of them and both reached for a rate or a share instead, which is what you do when the level is small enough to undercut the story.

None of that makes the disclosures dishonest. It makes them early, and early is where we are. The thing that changed this week is the type of evidence, not the size of the market. For most of a year the case for agentic commerce rested on people with a model. It now rests, partly, on two companies willing to be held to what they said. That is a real upgrade and it is worth marking.

Build accordingly. The acquisition signal in Finkelstein's second number is the part likely to still be true in eighteen months, so that is the one to plan against. Do not staff a team against a tripling.

A rate tells you something is growing. A share tells you nobody has shown up to compete yet. Neither one tells you how big it is, and this week both companies had the choice.