Agentic commerce is what happens when AI agents do the shopping. An agent takes over the buying, some of it or all of it... it searches, compares, decides, checks out, and pays on a person's behalf.
Exactly how much of the buying counts is itself contested. In the strictest read, agentic commerce is only the full loop... a query that ends in a completed purchase, no human in the middle. In a looser read, it already includes the assisted edges: an LLM surfacing products, comparing options, or handing a shortlist back to a person who makes the final call. That line, between agent-assisted and agent-completed, is one of the real open disagreements in the space, and different players draw it in different places.
Either way, it splits into a handful of very different problems. Getting discovered by an agent has little to do with getting checked out by one, and neither looks like moving the money. Each has its own players, standards, and open questions.
The shape of the space
When you evaluate agentic commerce as an actual journey, you get the individual pieces that make up its entirety:
- Discovery. The agent has to find your product before anything else can happen. Being surfaced and cited by AI systems is what matters here, across the many different queries an agent runs that are relevant to what you sell. It is a wider surface than a single blue link in Google, and it is where SEO turns into something new.
- Selection. Out of what surfaced, one product gets chosen. If the shopper makes that call, it looks like a normal shopping journey. The layer that is new to agentic commerce is when the agent chooses on a query's behalf... deciding which product is most relevant, or most deserving of the pick, with its own ranking (position, structured data, the model's taste) settling the winner.
- Readiness. Whether an agent can access your page, see your inventory, understand your product data, and navigate the site well enough to finish the task. In practice this is the first gate to participating in agentic commerce as a seller... if an agent cannot read and act on your store, none of the rest can happen.
- Checkout and payment. How the order gets placed, how the money moves, who is merchant of record, and how disputes, refunds, and liability are handled.
Running underneath all of it are the protocols, the standards different companies are proposing so agents and merchants can transact without a custom integration for every pair.
Each of those is its own discipline.
The stack in action
A shopper asks, the agent surfaces and ranks, the shopper picks, and the agent finishes the order. Open any step to see what happened, and go deeper.
Discovery works differently on each agent: open-web search over HTTP, a proprietary shopping index like Google's Shopping Graph, merchant product feeds, or product data exposed through a protocol like ACP or UCP. Whether a product surfaces at all, and where it ranks, is set by each model's own behavior and stays largely opaque.
Go deeper: Discovery · How Agents Buy
Readiness is whether an agent can actually transact here: the shoe is in stock at $190, and a machine can reach checkout without a login wall or a bot block in the way. On a UCP/ACP merchant this arrives as structured data; otherwise the agent parses it from the rendered page over HTTP.
Go deeper
Each stage above has a full guide, and the protocols underneath have one too:
Getting found by AI, and the gap between being indexed and being retrieved.
The behavioral layer, and why the pick changes from one model to the next.
What it takes for an agent to reach, read, and buy from your store.
How the money and the authorization move, and who is merchant of record.
UCP, ACP, MCP, and WebMCP (and HTTP), sorted by what each one does.
Keeping current
The Desk has the latest read on what's shipping and what it means.