TL;DR

• The agent is taking over the errand, not the relationship. Even OpenAI handed checkout back to retailers, and shoppers sent by AI arrive more engaged, not less.
• That gives the brand's own experience two new jobs: writing the shortlist before the agent is ever asked, and learning the context no catalog feed can carry.
• The opportunity is right there. While we all race to become readable by machines, almost no one is redesigning their experience for the better-informed visitor agents are now sending.

OpenAI built Instant Checkout so none of us would ever need to leave ChatGPT to buy something. Six months later, it handed the checkout back to the retailers, letting them steer the buying experience either on their own site or through an app inside ChatGPT that they control [1]. And yet, this September, in the very same industry, Mastercard, Visa and Instacart all doubled down on agents that search, choose and pay on our behalf. So which one is it? Are agents about to make the brand's own experience irrelevant, or are they quietly proving why it matters more than ever?

I've been following this closely since March, when I wrote about OpenAI dropping Instant Checkout (here), and the signals have kept piling up on both sides ever since. This is where they've taken me: the agent is taking over the errand, not the relationship. And that gives the brand's own experience a new job. Two, to be exact. (Full disclosure, this is literally my day job, so I know I have a bias. I've tried to let the data argue with me.)

First, the case that brands are becoming a feed

I want to give the other side a fair hearing, because a lot of it is right, and it got louder this month.

In the span of two days in September, Mastercard introduced Agent Connect, a single integration linking merchants, AI agents, platforms and payment providers [5]. Instacart launched Clementine, an assistant that builds a full grocery cart from one message [6]. Visa, Mastercard and Ant International announced work on common standards to verify the agents that buy on our behalf [7]. That isn't a chatbot feature anymore. That's infrastructure being poured.

McKinsey has a name for what this does to the journey: the compression of the funnel. Search, comparison and consideration collapse into a single moment the agent handles, and loyalty, as they put it, becomes less about sentiment and more about policy [8]. For everyday categories their verdict is even clearer. Brand storytelling and front-end experience matter less than operational trust, because apparently agents optimize for price, availability, reliability and easy returns.

Then there's the research on how AI models actually pick brands. In one study testing skincare recommendations, well-known brands were chosen 100% of the time when all products had the same specs. That dominance disappeared once a competitor had a rating advantage of less than 0.1 stars [9]. Trust me, that is not a comfortable number to read when you've spent most of your career in brand building. And we marketers already feel it, don't we? According to Braze, 71% of marketing leaders say agents have already weakened their ability to connect directly with customers [10].

So let me say it plainly. For detergent, for the weekly shop, for anything we buy on repeat without a second thought, the agent will most probably do a better job than any homepage we could design. (Sounds like interesting times for my former colleagues at Unilever and Nestlé.) I'm not going to pretend otherwise.

And yet, the agents keep sending people home

Here is where it gets interesting, and it brings us back to OpenAI.

The official explanation was about flexibility. OpenAI said the first version of Instant Checkout didn't give merchants the flexibility it wanted, so it would let them use their own checkouts while it concentrated on product discovery [1]. Analysts were less diplomatic. They told CNBC that OpenAI had underestimated how hard transactions are: onboarding merchants, keeping product data accurate, supporting multi-item carts and connecting loyalty memberships all proved difficult [3]. Walmart's Daniel Danker had already seen it coming. Speaking at a Morgan Stanley conference in early March, he called Instant Checkout "a very temporary moment in time," and said Walmart's own assistant, Sparky, would be the experience that travels into ChatGPT and Gemini instead [2]. The numbers behind that confidence were telling: checkout completed inside ChatGPT reportedly converted at about a third of the rate of sending the same shoppers to Walmart to finish the purchase [4].

Look at what Walmart built in its place. It is a ChatGPT app that carries shoppers from discovery into a Walmart environment, complete with account linking, loyalty and Walmart payments [1]. In other words, when the most powerful AI platform in the world tried to own the whole journey, the journey found its way back to the brand.

And it isn't only Walmart. Adobe's July data shows shoppers arriving at retail sites from AI platforms spend 59% more time on site, bounce 33% less, add to cart 28% more and convert 60% higher than non-AI traffic [11]. (Adobe also sells a product that helps brands attract more of that traffic, so I'd take it with the right pinch of salt. The direction is still hard to ignore.) Those aren't visitors the brand has lost to the agent. They're better ones.

The shopper side tells the same story. commercetools finds 65% of consumers trust AI to compare options, but only 14% trust it to make the purchase [12]. Visa's Trust Index lands in a similar place: 72% of US adults have used an AI assistant, yet only 23% would trust generative AI itself to execute a payment [13]. People are happy to let AI inform the decision. They're just not letting it make the decision. 

Not yet, anyway.

And retailers are acting on it. Instead of handing the relationship to someone else's agent, they're bringing agents inside their own walls. Instacart's Cart Assistant lets grocers run their own branded agent in their own app, built on the retailer's catalog and customer data [6]. Amazon says shoppers who use Rufus are 60% more likely to buy [14].

So what is actually happening? The agent runs the errand. Yet the person still comes home to the brand. The difference is that they now arrive knowing more, expecting more, and with far less patience for an experience that treats them like a first-time browser.

Two jobs no agent can take off a brand's hands

The way I see it, this leaves the brand's own experience with two jobs, and neither of them can be delegated.

The first is writing the shortlist before the agent is even asked. McKinsey's research in Europe makes a point I keep coming back to: loyalty is unlikely to disappear, but it may be expressed differently, with consumers effectively handing their agents "allow lists" of brands they trust and brands they want excluded [15]. So where does that list get written? Not inside the agent, that's for sure. It gets written over hundreds of small moments long before it, like a story someone tapped through on their commute, a quiz that told them something about their own taste, a recommendation that actually landed, or a return that went better than expected. None of that happens in a chatbot. It happens in experiences the brand owns. McKinsey also notes a ceiling to delegation: in some categories people will deliberately stop short of full automation, not because the agent can't do it, but because being involved is part of the value itself [8]. Choosing a gift, a look, a trip, searching for a prom dress, a first home. Nobody wants to outsource the part that feels like them. This is the part I am deliberately insistent on, because as a consumer, as an end user, today I do feel this. There is a dopamine effect here that none of us calculate properly. Handing it over fully to an agent doesn’t seem plausible for the foreseeable future for me, at least that’d be my two cents.

The second job is learning what only the relationship can learn. Agents run on structured information, and McKinsey warns that products which make perfect sense to people but can't be read by machines risk simply becoming invisible [8]. The obvious response is better product data, and yes, we should all be doing it. But specs are table stakes, and every competitor will have them too. Remember the 0.1 stars? What actually sets a brand apart is context: what this particular person asked, picked, skipped, saved or voted for. That kind of context only exists where the brand and the customer spend time together, and it is exactly what a brand's own agent needs to be any good. It's no coincidence that both Walmart's Sparky and Instacart's Cart Assistant are built on the retailer's own customer relationship. Agents read catalogs. People, on the other hand, remember experiences.

The opportunity sitting right there

Here's the part I find genuinely exciting. Right now, most of us are racing to become readable by machines: cleaning up product feeds, adopting new protocols, figuring out what generative engine optimization means for our category. All necessary, no argument there. But almost no one is redesigning their own experience for the visitor the agents are now sending.

Think about who that visitor is. They already know the specs. They've compared the prices and read the reviews, or at least a very confident summary of them. What they come for is depth, taste and a reason to care. And most of us still greet them with the same, lifeless banner carousel we built for someone who knows nothing yet.

So the generous move is to give them the stage, for once. Ask them instead of telling them. Let them vote between two looks, show them a video from their community to relate with, pick a mood or share what they're cooking this week, and then show them something that reflects it back. (This is the part I get to see up close at Storyly, when brands publish content that clicks in formats that move the thumb intuitively within their app. They aren't only getting a tap, they're getting an answer, experiencing a micro-dopamine moment. And an answer is exactly the kind of context no catalog feed will ever carry.) That one moment does two things at once. It makes today's visit better for the person, and it makes the brand's own agent a little smarter for tomorrow.

Where I've landed, for now

I say "for now" on purpose, because this topic is moving every single week, and I fully expect the signals to keep coming. So I want to stay hungry and humble. But here's where it stands today.

The announcements this September were all about rails: integrations, protocols, verification standards, all so agents can move through commerce safely. The question they leave us with is a different one. It isn't how we get agents to our products. It's what we build for the people once they arrive. Back in April, I wrote that personalization is only as powerful as the experience it lives inside. Agents don't change that. If anything, they make it truer, because an agent is only as good as what it knows about the person it's shopping for, and most of that is learned in experiences the brand owns.

I'm not saying I have the full playbook for this. But if an agent could handle the task, which of our experiences would a customer still choose to open? That's the question I'll be putting on the table for every app we work with this quarter. And if you've spotted a signal I've missed, in either direction, I'd genuinely love to hear it.

References

1. OpenAI, "Powering product discovery in ChatGPT," March 2026. openai.com
2. CNBC, "OpenAI's first crack at online shopping stumbled. It's preparing for the next wave," 20 March 2026. cnbc.com
3. CNBC, "OpenAI revamps shopping experience in ChatGPT after struggling with Instant Checkout offering," 24 March 2026. cnbc.com
4. Digital Applied, summary of Walmart's in-ChatGPT conversion data, 2026. digitalapplied.com
5. The Paypers, "Mastercard launches Agent Connect for agentic commerce," September 2026. thepaypers.com
6. Instacart, "Meet Clementine," press release, September 2026. instacart.com
7. Reuters via The Star, "Payment firms Visa, Mastercard and Ant International team up on AI agent trust framework," 10 September 2026. thestar.com.my
8. McKinsey, "The automation curve in agentic commerce." mckinsey.com
9. Chu et al., "Incumbent Advantage: Brand Bias in LLM Recommendation Systems," arXiv preprint, 2026. arxiv.org
10. Braze, "Agentic commerce." braze.com
11. Digital Commerce 360, "Adobe AI referral traffic data, July 2026," August 2026. digitalcommerce360.com
12. commercetools, "The Agentic Commerce Radar." commercetools.com
13. Crowdfund Insider, "Visa research: consumer trust is enabling path to agentic commerce," September 2026. crowdfundinsider.com
14. eMarketer, "Walmart and Amazon make different bets on AI." emarketer.com
15. McKinsey, "Europe's agentic commerce moment: decision influence is here, execution is coming." mckinsey.com