If you have not read the McKinsey piece on the “agentic commerce opportunity” yet, it is worth a skim because it captures the core shift in one sentence: shopping is moving from humans clicking around websites to software agents doing the legwork, and sometimes completing the purchase too.
[Mckinsey Report: The agentic commerce opportunity: How AI agents are ushering in a new era for consumers and merchants. Oct 2025. ](The agentic commerce opportunity: How AI agents are ushering in a new era for consumers and merchants)
The practical bit: what does that mean for brands, retailers, and anyone who has spent the last decade optimising product pages, ads, and checkout funnels?
It means we are entering an era where the “shopper” is increasingly an interface layer, a model, or an agent sitting between you and the person with the money. Your job is no longer just to persuade a human. It is to make your offer legible to machines, purchasable by machines, and still loveable to humans.
Let’s break it down.
**What “agentic commerce” actually is**
Agentic commerce is when an AI system does more than recommend. It can:
- understand intent (“I need a carry on suitcase that fits EasyJet rules and will not fall apart”)
- compare options across merchants
- ask follow-up questions
- negotiate or optimise (price, delivery speed, warranty)
- execute steps (build baskets, apply vouchers, check out)
- manage post purchase tasks (tracking, returns, replenishment)
The consumer is still in charge in theory, but in practice a lot of decisions get delegated into “micro-decisions” the agent makes on their behalf.
You can already see the conditions forming:
- Google says AI Overviews in Search has over 1.5 billion users per month.
- OpenAI’s CEO said ChatGPT hit 800 million weekly active users (Oct 2025).
- Adobe reported generative AI traffic to retail sites was up 4,700% year on year in July 2025.
That is the top of funnel changing shape in real time.
**The three ways agents will buy (and why you should care)**
You can think of agentic commerce in three interaction patterns. Each one changes where value is captured.
1) Agent to site
The agent comes to your site or app and behaves like a super-user: it searches, filters, compares, and checks out. This is the least disruptive version because you still “own” the environment, but you must make the site machine-friendly.
2) Agent to agent
Your customer’s agent talks to your brand’s agent (or your commerce stack’s agent) and they transact, potentially with negotiation, bundling, substitutions, and policy checks along the way.
3) Brokered agent to site
A platform sits in the middle and orchestrates. This is where the power shift gets spicy, because the broker controls routing, ranking, and sometimes the checkout surface.
If you are a retailer, the risk is obvious: disintermediation. BCG puts it bluntly, retailers can end up as “background utilities” if they do not intervene.
Discovery is becoming “zero click”, and that breaks old playbooks
Classic eCommerce growth assumed a journey like this:
Search → click → product page → cart → checkout → email flows → repeat
Agents compress that. Sometimes there is no click at all. The user asks a question, the agent answers, and the purchase happens inside the chat or assistant surface.
That is not theoretical anymore:
- OpenAI describes “buy it in ChatGPT” flows powered by an open standard it co-developed with Stripe.
- Visa launched “Visa Intelligent Commerce” to open its network to AI-driven “find and buy” experiences.
- Google just announced a new open standard for agent based shopping called the Universal Commerce Protocol (UCP).
- Shopify says UCP is co-developed with Google and that native shopping is rolling into Google surfaces, plus expanded Copilot checkout integration.
**So the question for brands is not “how do we rank?” It is “how do we get selected when an agent is choosing?”**
The new optimisation target: the agent experience
For the last decade, teams obsessed over customer experience (CX). In agentic commerce, you need CX and AX: agent experience.
AX is basically: can a machine confidently understand your offer, trust it, and complete a purchase without falling over?
Here is what that typically requires.
1) Product data that is unambiguous
Agents do not “browse” like humans. They ingest structured facts and look for consistency.
You want:
- clean titles (no fluff, no keyword soup)
- consistent attributes (size, colour, materials, compatibility)
- rich metadata (care instructions, compliance, certifications)
- clear variant logic (what changes, what does not)
- transparent shipping and returns
If your catalogue is messy, the agent will either ignore you or misrepresent you, and both outcomes are expensive.
2) Real-time availability and fulfilment confidence
Agents optimise for outcomes: “will this arrive before Friday?”, “is it actually in stock?”, “what happens if it does not fit?”
If your stock data is stale or your delivery promises are vague, you lose to the merchant that gives the agent certainty.
3) Purchase and post purchase actions as APIs, not pages
Even if an agent can “use a browser”, the market is clearly heading towards programmatic commerce flows.
That is why standards and rails are emerging:
Google’s Agent Payments Protocol (AP2) uses cryptographically signed “mandates” to tie user intent to a verifiable purchase instruction.
Stripe and OpenAI’s Agentic Commerce Protocol is pitched as a way to make checkouts “agent-ready” without rebuilding backends.
If your checkout is a maze of scripts, popups, and brittle steps, agents will struggle. And if agents struggle, they route elsewhere.
4) Loyalty, subscriptions, and personalisation that still work
Agents will get very good at value extraction: managing subscriptions, cancelling, swapping, price tracking, stacking rewards.
That is not a reason to panic, but it is a reason to make your loyalty proposition genuinely worthwhile, not just a flimsy points scheme held together by forgotten logins.
5) Trust, consent, and governance
As soon as you let software spend money, trust stops being a brand slogan and becomes infrastructure.
There is also a genuine risk angle. Research on agent-to-agent negotiations shows different agents can secure very different outcomes, and automation can introduce weird failure modes like overspending or agreeing to poor deals.
So, you need guardrails:
- explicit consent flows
- spend limits
- auditable logs
- fraud and bot detection that understands “good agents” vs “bad automation”
- clear dispute handling
Why all of this is accelerating right now
Two forces are colliding:
The interfaces are shifting
Google and OpenAI are both pushing commerce into conversational surfaces.
The capability curve is steep
METR’s research suggests the length of tasks AI can reliably complete (their “time horizon”) has been doubling roughly every seven months.
When agents can string together longer chains of actions reliably, they stop being “assistants” and start being “doers”.
**Where Lmo7 fits in (without the hard sell)**
At Lmo7 we focus on exactly this intersection: agentic eCommerce and visibility in AI discovery. We help brands win “AI shelf space” across systems like ChatGPT, Gemini, and Amazon Rufus, using an LLM visibility Framework and multi-model auditing so you can see how you show up, where you are missing, and what to change.
In plain English, we work on the stuff agents actually use to decide:
- structured product understanding (catalogue and content)
- credibility signals (consistency across the web)
- marketplace readiness (especially Amazon)
- monitoring and iteration as the interfaces evolve
If you are already feeling your organic traffic wobble, or you suspect your products are being summarised badly by AI, that is usually a sign your “agent experience” needs attention.
**The FAQs people keep asking**
Will SEO die?
Not die. It will change shape. You still need discoverability, but increasingly you are optimising for retrieval and selection inside AI systems, not just blue links.
Will agents make everything a price war?
They can increase price pressure, yes, because agents are ruthless comparators. The counter is differentiation that is machine-readable: guarantees, bundles, availability, fit guidance, service levels, trust.
Should we build our own agent?
Maybe, but it is rarely step one. Step one is making your commerce stack agent-compatible so third-party agents can transact cleanly, then deciding whether a proprietary agent adds genuine value.
What is the biggest mistake brands will make?
Treating this like a “marketing channel” only. It is a full-stack shift: data, payments, trust, fulfilment, and content all matter.