Why agentic commerce settles on retailer-controlled checkout by the 2026 holidays: 3 signals

The prediction is straightforward: by the 2026 US holiday season, and confirmed when fourth-quarter results land in February 2027, agentic commerce will run predominantly on a retailer-controlled model rather than a universal in-chat agent that holds the catalog and closes the sale. The likely shape is reroute-to-retailer: an AI assistant surfaces products, then hands the shopper to the retailer’s own site or a retailer-branded assistant app, with the retailer staying merchant of record. Recent signals point away from the tidy “one agent buys everything for you” narrative and toward a more fragmented, retailer-gated surface. The pattern suggests value accrues to first-party demand and retail media, not to agent intermediaries.

This is not a claim that in-chat checkout dies. It is a claim about where the center of gravity sits by Cyber Week 2026: control, catalog and the merchant-of-record relationship stay with retailers, and the assistant becomes a discovery and routing layer rather than the storefront. Three concrete moves in the last month sharpen that read.

In short

• Prediction: agentic commerce settles on a retailer-controlled, reroute-to-store model by holiday 2026, with retailers as merchant of record, not universal in-chat agent checkout. Falsifiable at Cyber Week (late November 2026) and in Q4 earnings (February 2027).
• Signal 1: Salesforce made Agentforce Commerce generally available on July 6, 2026, with catalog-synced retailer agents plugged into ChatGPT, keeping the merchant in control of the storefront.
• Signal 2: the live surface has shifted to retailer-owned ChatGPT apps and reroute flows (Sephora, Target, Walmart, Best Buy, Home Depot, Nordstrom) after OpenAI pulled in-chat Instant Checkout in March 2026.
• Signal 3: The Trade Desk named retail-media pioneer Kristi Argyilan as Chief Commercial Officer on July 15, 2026, a bet that commerce value pools around retailer demand and media, not agent middleware.
• Counter-signal: a scaled OpenAI checkout relaunch, network payment tokens, or a forced interoperability standard could reopen universal in-chat buying before year-end.

Why this matters now

For two years the assumed endgame was a single assistant that browses, compares and buys across every store while the human watches. That framing shaped roadmaps, investor decks and a wave of protocol launches. The 2026 evidence is more awkward: the assistant everyone assumed owned AI shopping is the one that pulled its in-chat checkout, and the quieter players kept routing shoppers back to retailer sites.

The distinction is not academic. Whoever owns the merchant-of-record relationship owns the customer data, the returns policy, the loyalty hook and the media inventory attached to the transaction. If the assistant merely reroutes, retailers keep those assets; if the assistant transacts in-chat, the assistant captures them. The next 90 days of holiday execution likely decide which model becomes the default that 2027 budgets are built around.

That is why the shape of the surface, not the raw growth of AI-referred traffic, is the variable to watch. AI-referred retail traffic grew sharply through the first half of 2026 and converts better than most channels, which is precisely why the control question now matters more than the adoption question. The debate has moved from “will shoppers use agents” to “who keeps the customer when they do.”

There is a timing reason to make this call now rather than in the calm of January. The holiday quarter is when retailers freeze systems, lock partner integrations and commit media budgets, so the surfaces that go live in November tend to set the default for the following year. Whatever configuration handles the 2026 peak is likely to harden into the baseline that 2027 roadmaps assume. A prediction made in July is therefore testable against a genuine commitment point, not a quiet stretch of the calendar.

Signal 1: Salesforce ships retailer-controlled agents into ChatGPT

On July 6, 2026, Salesforce made its Agentforce Commerce release generally available, positioning three agents (a Shopper Agent, a Buyer Agent and a Merchant Agent) with native integration into ChatGPT, and Google Search plus the Gemini app flagged for the following months. The mechanism matters more than the branding. Product catalogs sync straight from the retailer’s Business Manager into ChatGPT, with no third-party middleware sitting between the store and the assistant.

Read structurally, this is an infrastructure vote for retailer control. The retailer owns the catalog, the agent logic and the customer relationship; ChatGPT becomes a distribution surface, not the storefront. Salesforce paired the launch with a pointed adoption claim, that retailers running their own shopper agents grew sales markedly faster than those on the sidelines, which frames first-party agents, not third-party ones, as the growth story.

The significance is that a major commerce stack vendor is not building toward a neutral universal buyer. It is arming individual retailers to project their own agent into the assistant. That is the opposite of a single agent that flattens every store into an undifferentiated catalog. It echoes the argument that agentic commerce will not crown one standard in 2026 and instead runs through abstraction layers each retailer controls.

There is a competitive-dynamics reason this matters beyond one vendor. Commerce platform vendors sell to retailers, so their incentive is to make the retailer stronger inside the assistant, not to build a neutral buyer that would commoditize their customers. When Salesforce, and by extension the broader Shopify and commerce-stack ecosystem, ships tooling that keeps the catalog and the checkout on the retailer’s side, it is aligning a large installed base of merchants behind the reroute model. That alignment is hard to reverse mid-cycle.

What to verify

The falsifiable core here is the merchant-of-record question. If, by the holidays, Salesforce-powered agents complete purchases inside ChatGPT with Salesforce or OpenAI as merchant of record, the thesis weakens. If they continue to sync catalogs and route the buy back to retailer infrastructure, it holds. A second check is disclosure: watch whether commerce vendors start reporting agent-sourced GMV as retailer revenue rather than platform revenue, which would confirm where the transaction legally sits.

Signal 2: the live surface is retailer-owned apps, not in-chat checkout

The clearest tell is what actually shipped versus what was promised. OpenAI launched in-chat Instant Checkout in late 2025, reached only around 30 Shopify merchants, and pulled the in-chat checkout surface in March 2026. The pivot was not a retreat from commerce; it was a redirection toward retailer-owned apps inside ChatGPT.

By mid-2026 the named early partners for that model read like a mainstream retail roster: Target, Sephora, Nordstrom, Best Buy, The Home Depot and Walmart, with Walmart bringing its Sparky assistant into ChatGPT and Sephora launching a branded app inside the chat surface. Crucially, these apps route the shopper to the retailer rather than completing the purchase in the assistant. The retailer keeps the checkout, the data and the merchant-of-record status.

The rest of the live landscape rhymes with this. Google’s AI Mode reroutes shoppers to complete purchases on retailer sites via the retailer’s own payment flow, and Walmart plugged into that discovery surface in January 2026. Even Perplexity’s in-chat buy, which does complete in the interface, leaves the retailer as merchant of record on the back end. The independent tracker that scores retailers on whether AI agents can shop their store captures the same pattern: rerouting, not in-chat completion, is the dominant sustainable model.

The consistency across otherwise-rival platforms is the tell. Google, OpenAI and the commerce vendors do not coordinate, yet they have converged on the same division of labor, with the assistant handling discovery and the retailer handling the transaction. When independent actors arrive at the same architecture without collusion, it usually reflects an underlying constraint rather than a fashion. Here the constraint is simple: retailers will not hand over the customer relationship, and no single assistant is large enough to force them to during one holiday season.

The sunset detail is worth dwelling on because it is the single most informative data point in the set. A product with the reach of ChatGPT reached only around 30 merchants on its in-chat checkout before pulling it, which suggests the friction was on the merchant side, not the consumer side. Retailers appear to have been reluctant to hand the transaction, and the data attached to it, to the assistant. That reluctance is the mechanism behind the whole thesis: the gate is being built by retailers protecting their economics, not imposed from outside.

One counter-current deserves naming. Amazon has been reported to be preparing to open its marketplace to some external AI agents, a move we examined in the piece on why Amazon opens its marketplace to external AI agents before the 2026 holidays. That looks like a contradiction, but it is better read as the same instinct: Amazon setting the terms of access on its own turf, gating who gets in rather than surrendering the storefront. Control, not openness, is the common thread.

Signal 3: the retail-media executive move

On July 15, 2026, The Trade Desk announced the appointment of Kristi Argyilan as Chief Commercial Officer and Executive Vice President, effective July 27. Argyilan is one of the most recognized architects of retail media, having led advertising at Uber and, before that, held senior roles at Albertsons Media Collective and Roundel, Target’s retail-media arm. At The Trade Desk she is set to lead data partnerships spanning identity, measurement and retail media.

Executive hires at this level are 30-day to 60-day strategic tells, not vanity announcements. Placing a retail-media heavyweight at the commercial helm of an independent demand-side platform signals a bet that the money in commerce pools around retailer first-party data and media, not around a neutral shopping agent. It is a wager that the entity owning the shopper relationship, the retailer, monetizes the agent era, and that independent ad infrastructure wants to sit next to that value.

This maps directly onto the reroute thesis. If assistants push shoppers back to retailer surfaces, the retailer captures the first-party signal that powers retail-media targeting, and platforms like The Trade Desk want a share of that flow. We traced the same logic in the analysis of why retail media and the independent DSP layer converge before year-end 2026. The hire is a personnel-level confirmation of a structural bet.

The skeptical reading is that one appointment proves little, and that is fair on its own. The weight comes from consistency: a retail-media specialist is being handed the commercial mandate at exactly the moment the commerce surface is deciding who owns the shopper. Companies staff ahead of where they think the value is heading, and independent ad infrastructure choosing a retail-media leader over, say, a pure agent-partnerships executive is a directional vote. It says the near-term money is expected to sit with retailer demand and measurable media, not with a neutral shopping agent that intermediates it away.

What the pattern suggests

Stacked together, the three signals describe one direction of travel. A commerce stack vendor is arming retailers with their own agents, the live assistant surfaces are rerouting rather than transacting, and an ad-tech leader is staffing for a world where retailer demand and media win. None of these is a press release about the same event; they are independent moves in infrastructure, product and talent that happen to point the same way.

The synthesis is that agentic commerce is maturing into a control layer rather than a disintermediation layer. Retailers appear to be conceding discovery to the assistants while defending checkout, data and media. That is a rational trade: discovery is contestable and increasingly commoditized, whereas the transaction is where margin, loyalty and monetizable signal live.

It also reframes what “winning” agentic commerce means. The likely victors are not the assistants with the slickest in-chat buy button, but the retailers whose catalogs, structured data and branded agents make them the easiest destination for a reroute. That is why the discipline of answer-engine readiness has quietly become a merchandising priority, a shift consistent with our view that agentic checkout becomes a named sales channel before year-end 2026 even as the checkout itself stays on retailer rails.

To keep the prediction honest, it helps to lay out the plausible holiday outcomes and rough odds rather than assert a single future. The scenarios below are not equally likely, and the point of naming them is to make the call scoreable. The base case is retailer-controlled reroute as the dominant surface, but two alternative paths are live enough to weigh explicitly.

Notice that the hybrid path does not break the thesis; it bends it. Even if more purchases technically complete inside a chat window, the prediction holds as long as the retailer remains merchant of record and keeps the data. The thesis only fails cleanly in the disruption scenario, where the assistant becomes the store rather than the front door. That is the specific outcome a future observer should look for.

Wider context: payments rails reinforce the gate

The payment networks are building for delegated, scoped authority rather than an all-powerful buyer, which quietly reinforces the retailer-control thesis. Mastercard’s Agent Pay binds a tokenized credential to a specific agent, a specific merchant scope and a specific consent policy. Visa’s Intelligent Commerce approach similarly constrains agent-initiated payments to defined merchant and session scopes.

The design philosophy is telling. Neither network is building a token that lets one agent buy anything anywhere without merchant-level guardrails; both treat the merchant scope as a first-class constraint. That architecture makes a universal, unbounded in-chat buyer harder to ship at scale and a scoped, retailer-anchored transaction easier. The rails are being poured in the direction of control.

There is a subtler consequence for fraud and liability that pushes the same way. Scoped tokens exist so that a compromised or misbehaving agent cannot spend beyond its mandate, which means the merchant must define the mandate, and defining the mandate is an act of control. A universal buyer with an unbounded credential would concentrate liability in the assistant and its payment partner, an exposure neither the networks nor the retailers seem eager to underwrite this year. The safer path, and therefore the likelier one, keeps the retailer in the loop on every scoped authorization.

There is a caveat inside this context, addressed below: the same tokenization could eventually make in-chat checkout secure and frictionless enough that retailers feel safe opening up. For now, the near-term effect points toward gating, because scoped tokens map naturally onto retailer-by-retailer permissioning rather than open catalogs.

Implications for retailers, platforms and investors

For retailers, the practical takeaway is to treat the assistant as a new top-of-funnel channel and to defend the checkout. That means clean, structured product data an agent can parse, a branded assistant presence where it is offered, and a reroute path that does not leak margin. The retailers likely to win the holiday are the ones already investing in answer-engine readiness and first-party data capture.

For platforms and assistants, the message is that owning discovery is valuable but not the same as owning the transaction. The sustainable model in 2026 looks like a revenue share on referral and media, not a merchant-of-record land grab. Assistants that push too hard for in-chat checkout risk the same merchant resistance OpenAI encountered.

For investors, the signal is to weight exposure toward the control layer: retailers with strong first-party data, retail-media networks, and the ad and payments infrastructure that sits beside the transaction. The convergence of retail media and independent demand-side platforms is a direct read-through, and executive moves like the one at The Trade Desk are the kind of tell that precedes capital allocation. The talent flow toward commerce media echoes the broader wave of retail chief AI officer appointments cresting before the 2026 holidays.

There is also a defensive read for brands that sell through many surfaces. If discovery is being conceded to assistants, the risk is margin leakage on referral fees and a slow erosion of direct traffic, so the counter is to make the owned channel the obvious reroute destination rather than a grudging one. Brands that treat the assistant as a shop window and their own site as the till are positioned to benefit; brands that let a third party own both are the ones most exposed to disintermediation if the disruption scenario arrives.

Caveats: what could go wrong

The prediction is falsifiable, and several plausible developments would break it. The first is an OpenAI relaunch at scale. OpenAI has iterated quickly before, and a holiday-timed return of in-chat Instant Checkout, backed by the Agentic Commerce Protocol and a broad merchant roster, would push the center of gravity back toward in-chat completion.

The second is the payment-token path. If Visa Intelligent Commerce and Mastercard Agent Pay make in-chat checkout secure and low-friction enough, a major retailer could decide the convenience is worth ceding some control, and competitors would likely follow fast if it captured share. The third is a forced-interoperability standard: a widely adopted universal protocol could compel retailers to make catalogs transactable by any compliant agent, eroding the gate.

A fourth caveat sits with Amazon. Its Buy for Me and Rufus surface already reaches hundreds of millions of shoppers, and if Amazon scales in-app agentic buying aggressively, that is a form of assistant-owned checkout, even if it lives inside a walled garden rather than a neutral agent. The thesis is about the dominant model across the open market, so a single dominant walled buyer would complicate the read without fully refuting it.

The honest position is probabilistic. The signals point to retailer-controlled reroute as the most likely holiday default, but the surface is moving fast and the counter-scenarios are live. A future observer should score this at Cyber Week 2026 and against Q4 earnings language in February 2027.

Frequently asked questions

What exactly is being predicted, and by when?

That by the 2026 US holiday season, and confirmed in Q4 earnings in February 2027, agentic commerce runs mainly on a retailer-controlled model: assistants reroute shoppers to retailer sites or retailer-branded apps, with the retailer as merchant of record, rather than a universal in-chat agent completing purchases across every store.

How is this different from saying agentic commerce is growing?

Growth is not in dispute; AI-referred traffic is rising fast. The prediction is about structure, specifically who keeps the customer relationship and the checkout. The claim is that control stays with retailers, not that adoption slows.

Does this mean in-chat checkout is dead?

No. Perplexity completes some purchases in-chat, and OpenAI could relaunch. The claim is that reroute-to-retailer, not in-chat completion, is the dominant surface by the holidays, not that in-chat buying disappears.

Why does merchant of record matter so much?

Because it anchors customer data, returns, loyalty and the media inventory tied to the transaction. Whoever is merchant of record captures the assets that make commerce profitable, which is why retailers are defending it and assistants would prefer to hold it.

Isn’t Amazon opening to external agents a counter-example?

It reads that way, but gating who may access its marketplace is still an exercise of control. Amazon setting the terms of entry on its own turf is consistent with retailers keeping the upper hand rather than surrendering it.

What is the strongest argument against this prediction?

A scaled, low-friction OpenAI relaunch backed by network payment tokens. If in-chat checkout becomes secure and convenient enough, a large retailer might open up, and competitive pressure could pull others along, shifting the center of gravity back toward the assistant.

What should retailers do about it now?

Make product data clean and machine-readable, establish a branded assistant presence where offered, protect checkout margin on reroute, and invest in first-party data capture. The retailers best prepared for answer-engine discovery are the likeliest reroute destinations.

How will we know if the prediction was right?

Check two things at Cyber Week 2026 and in Q4 earnings: whether the dominant live surface is reroute-to-retailer or in-chat completion, and whether any major US retailer has made its full catalog openly transactable by any third-party agent. If reroute dominates and no full catalog is opened, the prediction holds.

Where do the payment networks fit?

Visa and Mastercard are building scoped, delegated tokens that bind an agent to specific merchants and consent rules. That architecture currently favors retailer-by-retailer gating, though the same tokens could later smooth the path to broader in-chat checkout.

For the primary announcement underpinning the third signal, see The Trade Desk’s investor release on the appointment via its investor relations newsroom.