Agentic Commerce: What Visa, Mastercard and Walmart Are Telling Investors

I'm writing this on August 21, 2026, one day after Walmart's earnings call. The closing remarks ran through the standard retail liturgy — value, convenience, speed — and then ended somewhere that would have sounded absurd on a retail call two years ago: "agentic experiences" (full transcript).

Agentic commerce — AI agents that browse, compare and buy on behalf of consumers — has stopped being a conference-keynote topic and become a standing item on earnings calls. I work inside a database of 253,129 earnings call transcripts covering 12,799 companies, so instead of collecting anecdotes I can measure the thing directly: how many companies say the phrase, since when, in which sectors, and who repeats it most.

The short version: 89 companies mentioned "agentic commerce" on a call in Q1 2024. By Q1 2026 it was 350. And the companies mentioning it most often aren't the model labs or the chatbot builders — they're the companies that process the payments.

From 89 companies to 350 in two years

Method first, briefly: each number below is the count of companies with at least one earnings call in that quarter matching a full-text search for the exact phrase "agentic commerce". Keyword matching makes these counts an upper bound — some matches are incidental co-occurrences, an analyst floating the term in a question that management never picks up. The level is fuzzy; the trend is not.

Two things stand out.

First, the acceleration is recent. 2024 was a slow grind — 89, 109, 105, 127. The step change comes in late 2025: 206 to 305 between Q3 and Q4, a 48% jump in a single quarter, followed by 350 in Q1 2026. That is the shape of a theme moving from "some companies mention this" to "IR departments keep a prepared paragraph about this." From Q1 2024 to Q1 2026 the count roughly quadrupled.

Second, Q3 2026 isn't finished and already sits at 210 companies through August 21 — more than the whole of Q3 2025, with over a month of calls still to come. Against the roughly 10,000 calls that land in the archive every quarter, the theme now touches about one reporting company in thirty.

The center of gravity is the payments stack

Rank companies by matching calls since the start of April 2026 and the list is lopsided:

• Fiserv — 6 calls
• Mastercard — 6 calls
• Walmart — 5 calls
• Twilio — 5 calls
• Visa — 5 calls
• Salesforce — 5 calls

Three of the six top mentioners — Fiserv, Mastercard, Visa — are payments companies. That concentration is the tell. When a theme is speculative, the loudest voices are vendors selling the enabling technology. When a theme threatens or promises actual revenue, the incumbents whose revenue it is start talking about it on every call they hold. The payments networks are not sketching a product roadmap; they are describing a defensive perimeter. The other three names round out the transaction chain: the merchant (Walmart), the communications layer (Twilio), the customer-data layer (Salesforce).

The Q2 2026 sector counts make the same point from a different angle: Information Technology 112 companies, Consumer Discretionary 89, Financials 83, Communication Services 59, Industrials 51, Consumer Staples 25. Tech builds the agents. Discretionary and staples own the shelves the agents will browse. Financials own the rails the agents will pay over. All three sides of the transaction are narrating the same theme to investors — which is exactly what you'd expect when nobody is sure yet who captures the margin.

"Incremental transactions" — the incumbents' pitch

Here is Mastercard on its July 30 call (transcript):

"Agentic Commerce is the next evolution in payments where the importance of security, transparency and control only increase. Agentic Commerce creates a significant opportunity for Mastercard. It leads to incremental transactions."

Three sentences, and each one is doing strategic work. "Next evolution" — not disruption, evolution: the rails stay, only the interface changes. "Security, transparency and control only increase" — the trust layer gets more important, and trust is precisely what a card network sells. "Incremental transactions" — agents don't cannibalize volume, they add volume.

Visa's framing on its July 28 call is nearly interchangeable, right down to the company it keeps (transcript):

"I want to touch on 2 areas where we are deploying our new way of working with great impact, stablecoin and agentic commerce."

Note the pairing. Stablecoins and agentic commerce are the two topics that carry an implicit "or we get bypassed" for card networks, and Visa presents both as areas where it is "deploying" — offense, not defense. Stablecoins are their own story; I tracked that narrative separately, and it peaked on earnings calls three quarters ago. Agentic commerce, per the table above, has not.

The through-line from the networks is consistent: we are the trust layer for agents, and agents mean more transactions, not fewer. That's the bull case. The open question they don't spell out — because it is the bear case — is what happens if the entity the consumer actually trusts is the agent, and the agent is indifferent about the rails underneath it. Every layer between consumer intent and settled payment is up for renegotiation when the consumer is software. The networks' answer, repeated across six calls each in under five months, is that they intend to be the layer the agents settle on.

A bootmaker in Wyoming

The most concrete picture of what this looks like below the enterprise tier came from Fiserv on May 28 (transcript). The pitch is democratizing agentic commerce for the roughly 900,000 merchants on the Clover platform, and the example was "a single location bootmaker in Wyoming": when "someone is on Gemini searching for cowboy boots, that merchant wants to be" in the set of options the agent actually surfaces.

That's the long-tail story, and I think it's underrated. Most agentic commerce discussion defaults to the giants. Fiserv's example inverts it: the merchants with the most to gain are the ones who were never going to win a search-ranking fight or an ad-budget fight. If agents query structured product and inventory data instead of scrolling ranked web pages, a bootmaker's discoverability problem stops being a marketing problem and becomes a data-plumbing problem — and Fiserv wants to sell the plumbing. For a payments processor, that's a genuinely new position in the stack: not just settling the transaction, but determining whether the merchant is visible to the buyer at all.

Walmart's version: experiences, not rails

Retailers frame the theme differently, and Walmart — 5 matching calls since April — is the clearest example. Yesterday's closing remarks described the agenda as "serving customers and members with everyday great value, exceptional convenience and speed, all while pushing our business model forward, diversifying our profit mix and leading in agentic experiences" (transcript).

Read the sentence the way an analyst would: "agentic experiences" sits in the same clause as "diversifying our profit mix" and "pushing our business model forward." For the networks, agentic commerce is incremental volume on existing rails. For Walmart, it's filed under changing what the business is. The transcript doesn't say whether that means Walmart building its own shopping agents, optimizing its catalog for other people's agents, or both — which is precisely the question I'd want asked on the next call.

One soft quarter — noise, or the plateau again?

Now the wrinkle in the table: Q2 2026 came in at 325 companies, down about 7% from Q1's 350 — the first down quarter since a minor wobble in Q3 2024.

One quarter is noise, and I don't want to over-read it. But it rhymes with something I've measured before: the broader agentic AI theme also flattened in Q2 2026 after a long run of growth. Two adjacent AI narratives cooling in the same quarter can mean two different things: companies are doing less, or companies have stopped getting credit for saying it and the language has migrated into boilerplate. Earnings call vocabulary saturates before activity does — once a theme is table stakes, flagging it stops being worth a prepared-remark minute.

The early Q3 number argues against a real fade: 210 companies in roughly seven weeks. I'll re-run the query when the quarter closes.

Track it yourself

Everything above is one query with different parameters. The search endpoint takes exact phrases in quotes, plus AND/OR and -negation:

curl 'https://earningscalls.dev/api/v1/search?q="agentic commerce"&type=transcripts&date_from=2026-04-01' \
  -H "X-API-Key: $KEY"

That returns every matching transcript since April 2026. Shift the date window quarter by quarter to rebuild the series in the table, filter by sector to reproduce the split, or compose the query to narrow the theme — "agentic commerce" AND checkout, or "agentic commerce" -pilot if you only want companies past the experimentation stage.

If you'd rather run a standing monitor than one-off queries, I wrote up the full pattern — quarterly loop, per-sector counts, first-mention detection — in Build an Earnings Call Theme Tracker in Python. The parameter reference lives in the docs, and if your agent should do the querying itself — fitting, given the topic — the same archive of 11.99 million speaker segments is exposed as an MCP server.

What the transcripts already make unambiguous: the payments incumbents have decided agentic commerce is happening, and they are spending their scarcest resource — prepared-remark minutes, quarter after quarter — convincing investors they will be the ones collecting the toll.

Track any theme across 253,129 earnings call transcripts — full-text search, sector splits and quarterly counts via the earningscalls.dev API.