Agentic commerce is the shift from people shopping on websites to AI agents shopping on their behalf. An agent here means software that can discover products, compare them, and complete a purchase for a person, often without that person ever loading a product page. In 2026 this stopped being a conference slide and became real infrastructure. It also had its first public failure. Both of those things are worth understanding, because the lesson sitting between them is the one that actually helps you plan.
In September 2025, OpenAI and Stripe launched Instant Checkout inside ChatGPT: buy a product without leaving the chat. It was the headline demo for the whole category. By March 2026 it was gone. OpenAI retired the feature after roughly five months. Around thirty Shopify merchants had integrated, and sales were close to zero. OpenAI pivoted to retailer-operated experiences instead, the "ChatGPT Apps" now used by Walmart, Etsy, Target, Instacart, and others.
That is not a story about agent shopping being fake. It is a story about the flashy layer moving faster than the market was ready for. The protocol underneath it, the Agentic Commerce Protocol (ACP, the open standard for how an agent completes a checkout), survived. The consumer product on top of it did not.
If your AI plan in 2025 was to wait for the winning checkout button, 2026 already showed you why that was the wrong thing to wait for.
Underneath the failed demo, the plumbing kept getting built. It is settling into layers rather than a single winner:
The takeaway is not that you need to memorize three acronyms. It is that no single company owns this, the standards are designed to work together, and how they fit is still being negotiated in real time.
The demand signal is real too. Adobe research found that 38% of consumers have used generative AI for shopping, and most of those who tried agent-led shopping said it improved the experience. Shopify reported that orders originating from AI-powered search grew sharply through 2025. McKinsey projects agentic commerce could drive up to $1 trillion in US retail revenue by 2030. Analysts now track agentic commerce as an established shift rather than a prediction. Treat the specific figures as directional rather than gospel. The direction is not in doubt.
Here is the part the vendor keynotes skip. When an agent does the shopping, your website design, your merchandising instincts, and your paid-search playbook all lose leverage. The agent does not see your homepage. It reads your data.
BCG has warned that retailers who are not ready risk being reduced to background utilities inside agent-controlled marketplaces: a row in someone else's result, picked or skipped based on how machine-readable your catalog is. Visibility is shifting from "can a shopper find you" to "can an agent parse you, trust you, and recommend you."
The honest answer is unglamorous, which is exactly why it is worth saying out loud. You do not win agentic commerce by picking a protocol early. You win it by doing the foundational work that every protocol depends on:
This is what agent commerce readiness means. Not a transformation project. Not a bet on which AI company wins. It is the specific, checkable work of making sure that when an agent goes looking for what you sell, it finds something clean enough to sell for you.
The retailers who spend the rest of 2026 chasing headlines will have very little to show for it. The ones who spend it fixing their product data foundation will be the ones agents can actually work with in 2027.
Related reading: Decoding Ecommerce Costs with Shopify and Salesforce.