OpenAI is killing Instant Checkout. Don’t dance on the grave of agentic shopping yet

I additionally wasn’t shocked by the victory laps all over the place. “Agentic commerce is useless.” “We informed you so.” The naysayers are having a area day.

I reckon they’re celebrating the unsuitable factor.

What truly died

In response to The Info, OpenAI is killing off Instant Checkout, the function that lets customers purchase merchandise with out leaving ChatGPT. Going ahead, if you wish to truly buy one thing ChatGPT recommends, you’ll both use a third-party app constructed inside ChatGPT (like Instacart or Expedia) or get bounced to the retailer’s personal web site to finish the transaction there.

Because it seems, product information throughout the web is simply too messy, unstandardized, and fragmented to permit an AI agent to reliably automate checkouts with out risking monetary errors or stock disputes. As of this week, OpenAI had solely built-in a dozen out of hundreds of thousands of Shopify retailers. It hadn’t constructed gross sales tax assortment. The Agentic Commerce Protocol survives in narrower kind, however the in-chat checkout layer is gone.

Marc Massar at Aura Labs put it properly in his essay this week: what died is a selected enterprise mannequin the place OpenAI tried to be each the pipe and the toll sales space, inserting itself as a brand new checkout middleman and taxing each transaction. Customers rejected the toll sales space. The market was proper to kill it.

However the commentary has made a class error. It’s treating one failed implementation as proof that all the behavioral shift is a mirage.

As a result of whereas OpenAI was fumbling checkout, agentic purchasing was already working in one other market. Alibaba’s Qwen app completes meals orders, journey bookings, and product purchases inside a single conversational interface – proper now, at scale. It really works as a result of Alibaba owns the AI mannequin, {the marketplace}, the cost rails, and the logistics. As Massar factors out, OpenAI tried to duplicate this with out proudly owning the stack. That’s what failed, not the idea.

So we all know the conduct can work. The query is how and when it arrives within the West.

Stalls in adoption occur

Now, earlier than the Qwen comparability will get anybody too excited, not each conduct that works in China catches on within the US, or not less than not on any predictable schedule.

In China, livestream purchasing emerged round 2016 and grew right into a market price roughly $807bn by 2024. Within the US, the primary wave flopped. Meta shut down Fb Stay Procuring in 2022. Amazon launched its TikTok-like short-form video and picture purchasing feed ‘Encourage’ in 2022 and killed it by 2025. The cheap conclusion on the time: American shoppers don’t need to store this fashion.

The stall lasted years. After which TikTok Store launched in September 2023 and located traction by embedding commerce inside leisure, creator belief, and native checkout. It labored as soon as it stopped behaving like “QVC on the web” and began feeling like a pure extension of an app folks already liked.

There’s a two-sided lesson right here. For the naysayers: a stall doesn’t imply the conduct is useless. For the fans (myself included): simply because Qwen proves the mannequin works in China doesn’t imply it’ll translate on to Western markets on any predictable schedule.

Shoppers don’t undertake new know-how as a result of it’s theoretically helpful. They undertake it when the product expertise is pure, worthwhile, and protected. ChatGPT’s model of agentic checkout ran headfirst into the tough operational actuality that retail achievement, cost processing, tax compliance, and fraud prevention are extraordinarily thorny issues. That’s an execution failure, not a conceptual one.

The threats to retailers that persist

I’ve spent the previous couple of months arguing that AI-enabled commerce poses an actual menace to the $60bn+ retail media business – that when discovery strikes upstream into AI assistants, the on-site promoting surfaces the place retailers earn 70-80% margins begin shedding their viewers. I’ve laid out that case intimately in earlier columns here and here. 

Nothing about this week modifications that thesis. Three issues are price flagging:

Discovery has already moved upstream. Shoppers are utilizing AI assistants for product analysis, comparability, and narrowing their consideration units proper now. This doesn’t require checkout to work inside ChatGPT. Capgemini analysis discovered that 53% of US shoppers have made a purchase order primarily based on AI suggestions. That discovery shift is already eroding the worth of retailer-owned surfaces – and it’s accelerating.

Amazon isn’t pulling again. It has every thing wanted to make agentic purchasing work: consumer context, the product graph, cost rails, achievement infrastructure, and Rufus getting extra succesful by the quarter. Massar makes an apt comparability to Alibaba’s Qwen app, which already completes purchases inside a conversational interface – as a result of Alibaba owns the AI mannequin, {the marketplace}, the funds, and the logistics. Amazon is the closest Western equal. And it is brazenly signaling curiosity in enabling purchases from different retailers by means of their AI instruments. That’s the true aggressive menace.

Google isn’t sitting nonetheless both. It has the purchasing graph, Service provider Middle, and is pushing Gemini and AI Mode aggressively. Google has fumbled in commerce earlier than – however the tempo of current improvement suggests a willingness to disrupt itself that we haven’t seen from it in years.

Russ Dieringer of the retail analysis agency Stratably posed an fascinating query in a LinkedIn thread this week: Did the $15bn Amazon funding in OpenAI have one thing to do with this pullback? That’s hypothesis, nevertheless it’s the form of query price sitting with. If OpenAI is now financially entangled with Amazon, its incentive to construct competing commerce infrastructure appears to be like relatively totally different.

The final 10% is more durable than the primary 90%

The bar for significant change in shopper conduct is excessive. My very own expertise confirms that: One absolutely agentic buy, by no means to return.

However right here’s what’s occurred since that first underwhelming checkout expertise: AI has develop into indispensable to each different stage of my shopping for journey. I exploit it continuously for product analysis, comparability, critiques synthesis, discovering offers, and determining what I want within the first place. That first 90% of the purchasing journey has already been rewired, not less than for me. And the buyer utilization information suggests I’m not alone.

The final 10% – the precise transaction – is the onerous half. There are a lot of technical hurdles to beat, together with order administration, product feeds, cost safety, and so forth. It additionally must be interesting to shoppers: loyalty program integration, dealing with returns and refunds, smarter replenishment than subscribe & save, price range optimization throughout my family spending. Cracking that code will take time.

However contemplate how Amazon’s personal 1-Click on ordering performed out. When it launched in 1997, shopping for one thing with a single button click on felt dangerous to lots of people. No affirmation web page? No overview step? It labored as a result of ultimately, consumers trusted the platform sufficient to let go of a well-recognized friction level. Amazon patented it, defended it aggressively, and it turned some of the consequential UX improvements in e-commerce historical past.

The “purchase now” button of the agentic future will want that very same mixture: a trusted platform, a solved operational backend, and an expertise that makes the previous approach really feel unnecessarily clunky. By the point somebody cracks it, we’ll all be so embedded in AI-assisted purchasing at each different stage that the ultimate step will really feel like the plain lacking piece relatively than a leap of religion.

The skeptics acquired the short-term name proper. This explicit model of agentic commerce was oversold and underbuilt. However are they proper in regards to the vacation spot, or the timeline?

The grave-dancing feels untimely. I’d hold your trainers on.