What is agentic commerce? A plain-English guide for business owners
Agentic commerce means AI agents finding, comparing and buying on behalf of people. Here's what it looks like in 2026 and why it changes how businesses get chosen.
“Agentic commerce” gets used loosely, so here’s the plain version. It’s commerce where an AI agent, working on behalf of a person, does some or all of the finding, comparing, choosing and buying. The person sets the goal and the constraints, and the agent does the legwork and, more and more often, the transaction itself.
There isn’t a single product behind the term. It describes a change in who your customer deals with first, and for a lot of purchases that first point of contact is becoming software.
What it looks like today
Here are a few examples from products that exist as of September 2026.
Meta’s Muse, launched September 8, can book movie tickets, schedule appointments and shop online for its owner. It remembers context across conversations, so “the usual place” means something to it.
Google’s Gemini Spark was announced at I/O in May and so far is available to paying Gemini subscribers in the US. It runs continuously on Google’s cloud rather than on your phone, and Google’s demos included it watching for price drops in the background. Purchases through the Agent Payments Protocol are scheduled to follow, with the owner approving each transaction.
ChatGPT has been doing product discovery for a while. OpenAI launched Instant Checkout in September 2025 so that people could buy inside the chat, with Etsy sellers and a handful of Shopify brands live and a million more Shopify merchants promised. Only a few dozen of those ever went live, and in March 2026 OpenAI moved checkout back to merchants’ own sites. Discovery and recommendation still happen inside ChatGPT, but the purchase happens on your site.
Those three have the biggest companies behind them. There will be more, and a lot of them will be built on the same handful of protocols.
The three jobs an agent does
It helps to separate what an agent is doing at each stage, because the work you have to do as a business is different for each one.
The first job is discovery, where the agent puts together a list of candidates. It draws on web search, maps, the platform’s own index, structured data on business websites, and whatever it has learned from past interactions. If those sources can’t make sense of your business, you’re not on the list.
The second is evaluation, where the agent narrows the list down. It checks whether you offer the thing, whether the price fits, whether you’re open, whether you serve the area, and whether the evidence (reviews, consistency, responsiveness) suggests that the transaction will go well. This is the stage where structured data and clear policies pay off.
The third is the transaction, where the agent books, orders or pays. Sometimes it does that on your site, sometimes through a protocol like ACP or UCP, and sometimes by handing the owner a link. The businesses that can complete this step without a human in the loop get chosen more, because the agent’s owner asked for an outcome and the agent is judged on delivering one.
Why this is different from search
Search puts you on a page with nine competitors and lets the person choose. An agent typically shows its owner one to three options, or just does the task, so there’s no page two.
That changes the economics of visibility, because ranking well enough to pick up some clicks stops being a strategy. You’re either in the small set that the agent presents, or you’re absent from a decision that the customer never even saw being made.
It also changes what “optimization” means. Search optimization is mostly about content and links. Agent optimization is about being machine-readable and transactable and about having a track record. Some of that overlaps with what SEO and GEO teams already do, but a lot of it doesn’t, and we go into the differences in GEO, AEO and agent optimization: what’s actually different.
The protocols, briefly
You don’t need to understand the protocols to get started, but you’ll hear the acronyms.
The Agentic Commerce Protocol (ACP) came from OpenAI and Stripe, and it’s what ChatGPT uses to read merchant catalogs and hand off purchases. The Universal Commerce Protocol (UCP) is Google’s. It launched at NRF in January 2026 and covers the whole journey from discovery to post-purchase, using a profile at a well-known URL on the merchant’s domain. Google’s Agent Payments Protocol handles the payment step for Gemini. Anthropic’s Model Context Protocol (MCP) is a general way for agents to talk to tools and data sources; UCP supports it as a transport, and Gemini Spark uses it to connect to outside tools.
If you’re on Shopify or a similar platform, support for these tends to arrive as a setting that you turn on, while on a custom stack each one is an integration project. We wrote a comparison of ACP and UCP for business owners who need to brief a developer.
What a business owner should take from this
We’d point to three things.
The shift is happening now, with products that have real users, and the platforms are moving quickly.
The work itself is concrete: structured data, readable policies, a transaction path that works without a human, consistent listings, and a reputation that holds up when a machine reads it. None of it requires you to bet on which agent wins.
And the advantage builds over time. Agents remember which businesses worked, so a first year of being easy to deal with builds a record that keeps getting you recommended.
If you want to know where your business stands today, request a free assessment. We check the same things an agent does and tell you what to fix first.
Find out what the assistants see when they look at your business
The assessment is free. A person checks where you stand in search and AI answers, runs the four agent checks (Find, Understand, Transact, Trust) and tries to book or buy, then emails you a one-page report with the three fixes we think matter most.