The 2026 holiday season is likely to be the first time agentic checkout, software agents that browse, choose and pay on a shopper’s behalf, is tested at meaningful consumer scale, with the verdict visible by the time December numbers land in January 2027. The signals from the last month point to a narrower outcome than the headline framing suggests. Agentic transactions are likely to remain a low single-digit share of holiday e-commerce sales, while the durable 2026 win accrues to the infrastructure layer: the checkout protocols, the payment-network trust rails and a thin band of flagship merchants wired in early. The pattern suggests that 2026 is the plumbing year, not the volume year, and that the payment networks’ talk of “millions of shoppers” is better read as a positioning claim than a forecast of holiday throughput.
In short
- Prediction: agentic checkout faces its first mainstream consumer test during the 2026 holiday season (Black Friday through late December), but agent-completed purchases are likely to stay in the low single digits as a share of holiday e-commerce sales, with the result observable by January 2027.
- Signal 1: at Google I/O in mid-May 2026, Google launched the Universal Cart on top of the Universal Commerce Protocol (UCP) and the Agent Payments Protocol (AP2), co-developed with Shopify, Etsy, Wayfair, Target and Walmart, with flagship brands including Nike, Sephora, Ulta Beauty and Wayfair lined up.
- Signal 2: Visa and Mastercard have moved agentic payments from pilots to production rails, with Visa’s Trusted Agent Protocol, Mastercard’s Agent Pay and Verifiable Intent layer, and Visa publicly forecasting agent purchases at holiday scale in 2026.
- Signal 3: merchants are increasingly running dual protocol stacks (Google UCP alongside OpenAI and Stripe’s Agentic Commerce Protocol), while large retailers like Walmart hedge by embedding their own assistants inside ChatGPT and Gemini rather than relying on a single checkout standard.
- Counter-signal: unresolved liability, returns and trust questions, plus a wide gap between agent-assisted discovery and agent-completed payment, mean the volume case could slip a full season even as the infrastructure ships on schedule.
Why this matters now
For two years agentic commerce was a demo. The shift in the last month is that the three layers required for it to work at scale, a discovery surface, a checkout protocol and a payment rail that assigns trust and liability, have all reached production within weeks of each other. That convergence is what makes the 2026 holiday season the natural first test. The capability is arriving precisely as the highest-traffic shopping window of the year approaches.
The timing is not coincidental. Vendors ship commerce infrastructure in the spring and summer so merchants can integrate before the Black Friday code freeze, which for most large retailers falls in October. A protocol that lands in mid-May is a protocol built to be live for the holidays. The question is no longer whether the rails exist, but how many shoppers actually hand a purchase decision to an agent when it counts.
That distinction matters because the incentives of the firms making the loudest claims are not neutral. Payment networks benefit from framing agentic commerce as inevitable and imminent, because it positions their trust layers as essential toll booths. Reading their guidance as a market forecast rather than a positioning move is the most common analytical error in this space right now. The signals are real; the interpretation needs discipline.
Signal 1: Google’s Universal Cart and the protocol land grab
The freshest and most consequential signal arrived at Google I/O in mid-May 2026. Google introduced the Universal Cart, a persistent cart that follows a shopper across Search, Gemini, YouTube and Gmail, sitting on top of two open standards: the Universal Commerce Protocol, which structures how product data, cart state and checkout handoffs are exposed to agents, and the Agent Payments Protocol, which governs how an agent is authorized to pay. Google describes the rollout in its own commerce blog (see Google’s agentic shopping announcement for the platform’s framing).
The notable detail is the co-development list. UCP was built with Shopify, Etsy, Wayfair, Target and Walmart, and the early checkout experiences name brands including Nike, Sephora, Target, Ulta Beauty, Walmart and Wayfair, alongside Shopify merchants such as Fenty and Steve Madden. That is not a roster of experimental pilots. It is a coalition of merchants who collectively account for a large slice of US discretionary spend, signing up to a standard months before the holidays.
Shopify’s involvement is the tell that matters most. When a platform that powers a meaningful share of independent US e-commerce co-authors a checkout protocol, the protocol gains a long tail of merchants almost automatically, because integration becomes a platform feature rather than a per-merchant project. The same dynamic that let Shopify push cross-border selling tooling to thousands of stores at once applies to agentic checkout. Distribution is the moat, and Shopify is the distribution.
The strategic read is that Google is treating agentic checkout as a standards war rather than a product launch. By open-sourcing UCP and AP2 and recruiting the largest merchants as co-authors, Google is trying to make its protocol the default substrate, the way a dominant standard quietly wins by being everywhere before anyone decides. That ambition is real. Whether consumers show up to use it this holiday is a separate question, and the protocol’s existence does not answer it.
Signal 2: the payment networks move from pilot to production
The second signal is that Visa and Mastercard have shifted agentic payments out of the lab. According to reporting from American Banker, both networks expanded their agentic AI deployments through late 2025 and into 2026, moving from controlled pilots toward production rails. Visa introduced its Trusted Agent Protocol alongside more than ten partners, a mechanism that lets merchants distinguish a legitimate shopping agent from a malicious bot, which is the precondition for any merchant to accept agent-initiated orders at all.
Mastercard’s parallel move is Agent Pay, paired with a Verifiable Intent trust layer and early bank partnerships, plus a documented expansion of agentic payments into new markets including Hong Kong. The networks are building the same thing in slightly different shapes: a way to cryptographically establish that an agent is acting on a real consumer’s verified instruction, so that authorization, disputes and liability have somewhere to land. Without that layer, agentic checkout is uninsurable, and no large merchant will switch it on.
Visa has been explicit about the timeline, publicly forecasting that millions of consumers will use AI agents to complete purchases by the 2026 holiday season and characterizing 2025 as the last year shoppers routinely shopped unaided. That is a striking claim, and it is the one most worth scrutinizing. The trust rails are genuinely shipping, which is the verifiable part of the signal. The consumer-volume forecast attached to them is a marketing artifact, and the gap between the two is where the prediction lives.
The payments angle connects to a wider restructuring of who controls checkout, a theme explored in our analysis of European payments consolidation. Agentic rails are another front in the same contest: whoever owns the trust layer between shopper, agent and merchant owns a recurring position in every future transaction. That is why the networks are investing ahead of demand. The toll booth has to be built before the traffic arrives.
Signal 3: merchants hedge with dual protocols
The third signal is behavioral rather than announced, and it is the most revealing. Merchants are not betting on one protocol. They are increasingly running Google’s UCP alongside the Agentic Commerce Protocol that powers OpenAI’s Instant Checkout, which launched earlier in 2026 with Stripe as a payments partner and has since been packaged into commerce platforms including BigCommerce’s agentic suite. Running both is now a recognized tactic for capturing agent traffic from multiple surfaces at once.
Dual-stacking tells you something important: merchants do not yet know which agent ecosystem their customers will use, so they are buying optionality. That is rational, and it is also a sign of an immature market. When a channel has clearly won, integration consolidates. The fact that retailers are wiring up two or three protocols at once is evidence that 2026 is a year of hedged experimentation, not committed adoption.
The Walmart move sharpens the point. Rather than leaning on a third-party Instant Checkout flow, Walmart has signaled it will embed its own Sparky assistant directly inside ChatGPT and Gemini, keeping the customer relationship and the data on its side of the line. Large retailers with their own gravity have little interest in becoming an interchangeable SKU feed inside someone else’s agent. They will participate in the protocols while quietly working to own the shopper themselves.
That tension, between platforms that want to aggregate demand and large merchants that refuse to be disintermediated, is the structural friction that will cap agentic volume this year. It echoes the dynamic that played out when TikTok Shop pushed native in-app checkout: the platform wants the transaction inside its walls, and the biggest brands want the customer in their own database. Neither side fully wins in year one, and the standoff slows adoption.
What the pattern suggests
Put the three signals together and a clear shape emerges. The infrastructure is converging on schedule and ahead of demand, which is exactly what you would expect in the plumbing year of a new commerce channel. The discovery surfaces (Search, Gemini, ChatGPT), the checkout protocols (UCP, ACP, AP2) and the payment trust rails (Visa Trusted Agent Protocol, Mastercard Agent Pay) are all live or near-live before the holidays. The supply side is ready.
The demand side is the open variable, and the signals say less about it than the vendors imply. Co-development rosters, protocol launches and trust-layer announcements all describe capability, not usage. None of the three signals is a measurement of how many shoppers will actually delegate a holiday purchase to an agent. The honest synthesis is that 2026 delivers the rails and a credible first test, while the volume verdict stays genuinely uncertain until the season runs.
The base case, then, is a successful infrastructure year with modest consumer throughput. Agent-completed purchases likely land in the low single digits as a share of holiday e-commerce sales, concentrated in the flagship merchants and the repeat-purchase categories where delegation is lowest risk: consumables, replenishment, known SKUs. The early value accrues to the rails and the platforms, not to a broad consumer behavior shift. That is a real milestone, but a quieter one than “millions of shoppers” suggests.
It helps to hold three scenarios side by side rather than anchoring on a single number, because the outcome hinges on consumer behavior that has not been observed at scale yet. The table below frames the range, with the base case as the most probable and the two tails as genuine, not symmetric, possibilities. The bull case requires a frictionless default to emerge inside a daily-use surface; the bear case requires only that trust friction and a liability scare behave the way they did for voice shopping.
| Scenario | Holiday 2026 agent-completed share | What would have to be true | Primary beneficiary |
|---|---|---|---|
| Bear | Near zero, discovery only | Liability scare or hard discovery-to-checkout gap; consumers research with agents but pay themselves | No one captures volume; rails still ship |
| Base (predicted) | Low single digits | Flagship merchants live, repeat-purchase baskets delegated, broad gifting stays manual | Payment trust rails and platforms |
| Bull | Mid single digits or higher | Frictionless default inside Search, Gemini or ChatGPT; one-tap-style behavior shift | Dominant agent surface plus its payment rail |
| Signal | Source (text-cited) | Date | What it proves | What it does not prove |
|---|---|---|---|---|
| Universal Cart + UCP + AP2 | Google I/O, Google commerce blog | Mid-May 2026 | Major-merchant coalition and discovery surface are live before holidays | That shoppers will use the cart at scale |
| Trusted Agent Protocol, Agent Pay | Visa, Mastercard, per American Banker | Late 2025 to 2026 | Payment trust and liability rails are in production | That holiday agent volume reaches “millions” |
| Dual UCP and ACP merchant stacks | Platform integration patterns, BigCommerce, Stripe | Early to mid 2026 | Merchants are buying optionality across ecosystems | That any single ecosystem has won |
Wider context: what first holidays look like for new channels
The most useful way to calibrate the 2026 prediction is to look at how prior commerce channels behaved in their first mainstream holiday. New checkout behaviors almost never arrive as a step change. They show up as a low single-digit share that compounds over several seasons, and the firms that win are usually the infrastructure providers who get paid regardless of which brand the consumer ultimately picks.
Mobile commerce is the cleanest precedent. Smartphones were ubiquitous years before mobile crossed a meaningful share of holiday transactions, held back not by capability but by checkout friction and trust. Buy-now-pay-later followed a similar curve: available widely before it became a default, then compounding once it was embedded at the point of sale, a dynamic still visible in how BNPL shapes discretionary spend today. Voice shopping is the cautionary case. The capability shipped, the volume never came, because consumers would not trust an unseen interface with discretionary purchases.
Agentic checkout sits somewhere between the BNPL curve and the voice-shopping disappointment, and which one it resembles is the central uncertainty. The bull case is that agents inherit the trust mobile and BNPL eventually earned. The bear case is that delegating a purchase to software feels more like talking to a speaker than tapping a phone, and the trust gap proves stickier than the vendors expect.
| Channel | Capability widely available | First mainstream holiday share (approximate) | Who captured early value |
|---|---|---|---|
| Mobile commerce | Early 2010s | Low single digits, then compounding | Payment processors, platforms |
| Buy-now-pay-later | Late 2010s | Low single digits at point of sale | BNPL providers, networks |
| Voice shopping | Late 2010s | Negligible, never compounded | Effectively no one |
| Agentic checkout | Mid 2026 | Low single digits (predicted) | Protocols, payment trust rails |
Implications for retailers, brands, platforms and investors
For retailers, the practical move this season is to treat agentic checkout as an infrastructure investment with an option attached, not a demand channel to forecast against. Wiring up UCP and ACP is cheap relative to the cost of being invisible to agents if adoption surprises on the upside, and the integration largely rides on existing platform tooling. The expected return this year is low, but the cost of absence is asymmetric, which makes participation the correct call even under the modest base case.
For brands, the priority is product-feed hygiene and the same conversion discipline that governs any new surface. Agents read structured product data, so the catalogs that are cleanest, best-attributed and most accurately priced will be the ones agents surface and select. The work resembles the unglamorous conversion plumbing behind recovering abandoned carts: it pays off only if the rest of the funnel is sound, and it rewards preparation over spectacle.
For platforms, the contest is about becoming the default agent surface and the default checkout standard, and the next two seasons will likely decide it. Google, OpenAI and the payment networks are each trying to occupy a different layer of the same stack, and the merchant dual-stacking behavior tells you none has won yet. The platform that converts protocol adoption into actual shopper habit first will compound the lead quickly, because commerce standards tend toward winner-take-most once a default emerges.
For investors, the cleaner exposure this cycle is the infrastructure layer rather than any single consumer surface. The payment networks and platforms get paid on agentic transactions regardless of which brand wins the basket, which is the toll-booth position that historically captures the durable value in commerce shifts. The riskier bet is on near-term consumer volume, where the base case is modest and the disappointment scenario is real. Owning the rails is the lower-variance way to be early.
Caveats: what could go wrong
The prediction could be wrong in either direction, and the counter-signals deserve honest weight. The strongest argument against even modest volume is unresolved liability. When an agent buys the wrong item, ships to the wrong address or triggers a fraudulent charge, the question of who eats the cost, the consumer, the merchant, the network or the agent provider, is not fully settled. Trust layers like Verifiable Intent are attempts to answer it, but they are new and untested at holiday volume, and one high-profile failure could chill merchant willingness fast.
The second caveat is the discovery-to-checkout gap. The most likely behavior this year is that shoppers use agents to research and compare, then complete the purchase themselves on a familiar surface. If that pattern dominates, agent-assisted discovery could be large while agent-completed checkout stays negligible, which would make the prediction’s “low single digits” look generous rather than conservative. Discovery and payment are different acts of trust, and the gap between them has sunk new channels before.
The third caveat runs the other way. If a major platform makes agentic checkout the frictionless default inside a surface consumers already use daily, adoption could outrun the base case, the way one-tap checkout once compressed mobile friction overnight. The Universal Cart living natively inside Search and Gmail is exactly the kind of distribution that can move a behavior faster than precedent suggests. The prediction assumes friction and habit win this season; a sufficiently seamless default could break that assumption.
The fourth caveat is regulatory. Consumer-protection regulators in the US and EU are already examining how AI agents handle disclosure, consent and dark patterns, and an enforcement signal or a pointed consultation during the second half of 2026 could make merchants cautious about switching agentic checkout fully on for the holidays. Regulatory drag would not kill the channel, but it could push the volume verdict into 2027, leaving 2026 as a pure infrastructure year with even less consumer throughput than the base case.
Frequently asked questions
What exactly is being predicted, and how can it be checked?
The prediction is that the 2026 holiday season becomes the first mainstream test of agentic checkout, but that agent-completed purchases stay in the low single digits as a share of holiday e-commerce sales. It is checkable in January 2027 when holiday e-commerce data and any platform or network disclosures of agentic transaction share are published. A reading clearly above mid-single digits would falsify the cautious volume call; near-zero completed volume would falsify it in the other direction.
Why not just trust Visa’s forecast of millions of shoppers?
Because the firms making the loudest adoption claims benefit directly from agentic commerce being seen as imminent. Payment networks earn a position in every agentic transaction, so framing the shift as inevitable is good positioning. The trust rails they are shipping are verifiable and real; the consumer-volume number attached to them is a marketing artifact, and the two should be assessed separately.
Is agentic checkout the same as AI-assisted shopping?
No, and the distinction is the heart of the prediction. AI-assisted shopping, where an agent helps you research and compare, is likely to be widespread this holiday. Agentic checkout, where the agent actually completes the payment on your behalf, is the higher-trust act that is likely to stay small. Conflating the two is the most common way to overstate near-term adoption.
Which merchants are best positioned to benefit?
The flagship merchants in the UCP and ACP co-development rosters, large retailers with clean product feeds and repeat-purchase categories like consumables and replenishment. Delegation is lowest risk when the item is known and the price is predictable, so those are the baskets most likely to be handed to an agent first. Discretionary gifting, where choice carries emotional weight, is likely to lag.
Should a smaller retailer integrate agentic checkout now?
Most likely yes, but as an option rather than a demand bet. If the retailer runs on a platform like Shopify or BigCommerce, the integration largely rides on existing tooling and is cheap relative to the asymmetric cost of being invisible to agents if adoption surprises. The realistic expectation is low return this year and durable positioning for later seasons.
What is the biggest risk to the prediction?
A frictionless default. If a platform embeds agentic checkout inside a surface consumers already use every day and removes enough friction, adoption could outrun the base case the way one-tap checkout once accelerated mobile commerce. The prediction assumes habit and trust friction win this season, and a sufficiently seamless default is the cleanest way that assumption breaks.
How does liability actually work when an agent buys something wrong?
It is not fully settled, which is precisely why the networks built trust layers like Verifiable Intent and the Trusted Agent Protocol. These aim to establish that an agent acted on a real consumer’s verified instruction, giving disputes and chargebacks somewhere to land. Whether they hold up at holiday volume is untested, and an early failure could make merchants cautious quickly.
Does Walmart embedding Sparky in ChatGPT contradict the protocol story?
It complicates it rather than contradicting it. Walmart is participating in the ecosystem while refusing to become an interchangeable feed inside someone else’s agent, keeping the customer relationship on its own side. That hedge by large merchants is exactly the structural friction that is likely to cap agentic volume this year, even as the protocols themselves succeed.
When will agentic checkout actually become a major channel, if not in 2026?
The precedent of mobile commerce and BNPL suggests a multi-season compounding curve rather than a single breakout, which would point to a more material share by the 2027 or 2028 holidays if trust and liability questions resolve. The 2026 base case is the plumbing year; the volume inflection, if it comes, likely lands one to two seasons later, assuming the channel does not follow the voice-shopping path into irrelevance.