What Is Agent Commerce? A Complete Guide for 2026

A year ago, the idea of an AI agent independently researching a product, comparing prices across retailers, and completing a purchase on your behalf was a research demo. Today, it is production infrastructure. Google, OpenAI, Stripe, Shopify, and dozens of major retailers have shipped protocols and integrations that make this real — and the adoption curve is steeper than most businesses realise.

This guide explains what agent commerce is, how it works under the hood, which protocols matter, and what your business needs to do to stay visible in a world where the buyer might not be human.

What is agent commerce?

Agent commerce — also called agentic commerce — is a model of electronic commerce in which AI agents act as intermediaries between consumers and businesses. Instead of a person browsing a website, clicking through product pages, and filling in checkout forms, an AI agent performs some or all of those steps autonomously.

The agent might be a general-purpose assistant (like those built on large language models), a specialised shopping agent, or an enterprise procurement system. What they share is the ability to discover products, evaluate them against criteria, negotiate terms, and execute transactions — all through machine-readable interfaces rather than visual web pages.

This is not a rebrand of chatbots or recommendation engines. Those tools assist a human who is still making every decision and clicking every button. In agent commerce, the agent itself takes actions: it reads structured data, calls APIs, compares options programmatically, and can complete a purchase without a browser ever rendering a page.

Why agent commerce matters now

The shift is not theoretical. The numbers tell a clear story.

McKinsey's 2023 research estimates that generative AI could unlock $2.6–4.4 trillion in annual value across industries — and agent-mediated commerce is positioned to capture a significant share. Gartner projects that AI agents will intermediate trillions in B2B spending by the end of the decade. And Deloitte's retail outlook research shows that a majority of retail executives plan to deploy agentic AI for key operations within the next two years.

The infrastructure is catching up to the ambition. Google's Shopping Graph already processes billions of product listings, and that data layer is now being exposed to AI agents through new open protocols. The question is no longer whether agent commerce will happen — it is whether your business will be part of it.

How AI agents discover, evaluate, and purchase

To understand why agent commerce requires changes to your website, it helps to understand the agent's workflow. It broadly follows three phases.

Phase 1: Discovery

An AI agent needs to find your business and understand what you sell. It does this by reading machine-readable signals: structured data markup (Schema.org), sitemaps, well-known endpoint files (/.well-known/ucp, agent.json, llms.txt), and API documentation. If your product information exists only as text on a rendered web page, most agents will struggle to parse it reliably — or skip you entirely.

Discovery is the most critical phase because it is binary. If an agent cannot find your products in a format it understands, nothing else matters. You are invisible.

Phase 2: Evaluation

Once an agent has discovered your products, it evaluates them against the consumer's criteria: price, availability, specifications, reviews, shipping terms, return policies. It does this programmatically — pulling data from structured fields, comparing across multiple retailers in parallel, and scoring options against weighted preferences.

This is where data quality separates winners from losers. An agent comparing running shoes across ten retailers will favour the listing with complete, accurate Schema.org markup — the one where price, size availability, colour options, ratings, and shipping estimates are all machine-readable. A listing with "call for price" or sizes buried in an image gets filtered out.

Phase 3: Purchase

The agent selects a product and completes the transaction. In the ACP model, this happens via API calls through Stripe's payment rails. In the UCP model, the agent negotiates through published commerce capabilities and pays through the Agent Payments Protocol. Either way, the checkout happens without a traditional browser session.

This breaks a fundamental assumption of most e-commerce platforms: that a human is present at checkout. CAPTCHAs, multi-step forms that require JavaScript rendering, and session-dependent carts all become barriers. Agent-ready checkout means API-accessible checkout.

The protocols powering agent commerce

Several protocols have emerged to standardise how agents interact with businesses. Here are the ones that matter most.

Google UCP (Universal Commerce Protocol)

UCP is an open protocol announced by Google at the National Retail Federation conference in January 2026. It covers the entire shopping journey — discovery, evaluation, negotiation, purchase, and post-purchase support.

Merchants publish their commerce capabilities at a /.well-known/ucp endpoint on their domain, similar to how robots.txt signals to search crawlers. Any AI agent — not just Google's — can discover these capabilities and negotiate transactions. UCP is backed by a coalition including Shopify, Stripe, Visa, Mastercard, Walmart, Target, and Best Buy.

UCP is designed to be protocol-agnostic. It works with REST APIs, MCP, and Google's Agent-to-Agent (A2A) framework, positioning it for a multi-agent future.

OpenAI/Stripe ACP (Agentic Commerce Protocol)

ACP is an open standard co-developed by OpenAI and Stripe. It focuses on enabling AI assistants to discover products, present options to buyers, and complete purchases within a conversational interface — currently ChatGPT.

ACP is deliberately focused on the checkout experience. Merchants who already use Stripe can integrate with minimal code changes. Shopify has announced ACP support across its platform, giving merchants access to ChatGPT's growing shopping audience.

MCP (Model Context Protocol)

MCP is an open-source protocol — originally developed by Anthropic and now governed by the Linux Foundation — for connecting AI models to external tools and data sources. In the commerce context, MCP allows agents to interact with business systems — inventory databases, pricing engines, order management platforms — through a standardised interface.

MCP is not a commerce-specific protocol, but it is becoming the connective tissue that lets agents access the business logic behind a storefront. UCP explicitly supports MCP as one of its transport layers.

agent.json

agent.json is a proposed convention (analogous to robots.txt) that lets websites declare their capabilities and preferences for AI agent interactions. Placed at the root of a domain, it tells agents what actions they can perform, what data is available, and what authentication is required.

Think of it as a handshake file: it does not enable commerce on its own, but it signals to agents that your site is ready for automated interaction and points them to the right endpoints.

llms.txt

llms.txt is a convention for providing AI-friendly content summaries at a site's root. Where robots.txt tells crawlers what they may access, llms.txt tells language models what your site is about and how to interpret its content. It is particularly useful for helping agents understand a business's products and services without having to parse every page.

The six dimensions of agent commerce readiness

Not all aspects of readiness are equal. At Zeodyn™, our scoring methodology evaluates websites across six weighted dimensions that collectively determine whether AI agents can successfully interact with a business. For a deep dive into how the scoring works — including real-world examples from major retailers — see The Zeodyn Score Explained.

1. Discovery & Access

Can agents find your site and its products? This dimension covers crawlability, sitemap quality, robots.txt configuration, well-known endpoints, and whether your content is accessible without JavaScript rendering. A site that blocks automated access or hides content behind client-side rendering fails at the first hurdle.

2. Structured Data

How complete and accurate is your machine-readable markup? Schema.org product markup, JSON-LD, Open Graph tags, and other structured data formats are the language agents use to understand your catalogue. Missing fields, outdated prices, or incomplete product attributes all reduce your score — and your chances of being selected.

3. Commerce Data

Beyond basic product information, agents need commerce-specific data: real-time pricing, stock levels, shipping options, return policies, payment methods. This dimension measures whether the data an agent needs to complete a transaction is actually available and machine-readable.

4. Protocol Support

Which agent commerce protocols does your site implement? UCP endpoints, ACP integration, MCP compatibility, agent.json, llms.txt — each protocol extends your reach to different agent ecosystems. Sites that support multiple protocols maximise their visibility.

5. Security & Trust

Agents need to verify that a site is legitimate before transacting. This dimension covers HTTPS configuration, certificate validity, security headers, Content Security Policy, and trust signals. An agent that cannot verify your site's security will not complete a purchase — and should not.

6. Technical Performance

Speed matters for agents just as it does for humans — arguably more, because agents operate at scale and will deprioritise slow responses. This dimension covers response times, server reliability, API latency, and whether your infrastructure can handle automated request patterns without degrading.

Who needs to care about agent commerce?

The short answer: any business that sells online. But some sectors face more immediate pressure than others.

Retailers and e-commerce businesses are on the front line. Agent commerce directly affects product discovery and sales. If your competitors are agent-ready and you are not, their products will be recommended and yours will not.

B2B suppliers and wholesalers face a similar dynamic. Enterprise procurement agents are already evaluating suppliers programmatically. As agent-mediated B2B spending grows, businesses that are not machine-readable will be invisible to procurement agents.

SaaS and service businesses may feel less urgency, but agents are increasingly used to evaluate and compare software, recommend service providers, and even initiate subscriptions. Structured pricing pages and API documentation matter here.

Marketplaces and platforms have a dual challenge: they need to be agent-ready themselves, and they need to help their sellers and merchants become agent-ready too. Shopify's early moves on both UCP and ACP reflect this — see our Shopify AI agent readiness guide for the specific steps merchants can take today.

Common misconceptions

"This is just SEO with a new name." Agent commerce builds on structured data practices that overlap with SEO, but it goes far beyond search rankings. Agents do not just find your products — they evaluate, negotiate, and purchase. The technical requirements (API-accessible checkout, protocol endpoints, real-time data) have no equivalent in traditional SEO.

"Only big retailers need to worry." The protocols are open standards. A small Shopify merchant implementing ACP gets the same access to ChatGPT's shopping interface as a multinational. In fact, smaller businesses that move early may gain disproportionate visibility while competitors wait.

"AI agents will replace my website." Agents do not replace websites — they add a new channel. Humans will continue to browse and buy directly. But agents will increasingly mediate discovery, and businesses that are invisible to agents will lose a growing share of traffic and revenue.

Key takeaways

• Agent commerce is a model where AI agents autonomously discover, evaluate, and purchase products on behalf of consumers and businesses.
• The market is large and growing fast. McKinsey, Gartner, and Deloitte all project substantial growth in AI-mediated transactions over the next two to four years, with trillions of dollars in value at stake.
• Multiple protocols are emerging. UCP (Google), ACP (OpenAI/Stripe), MCP (Linux Foundation), agent.json, and llms.txt each serve different parts of the agent interaction lifecycle. Supporting multiple protocols maximises reach.
• Readiness spans six dimensions: Discovery & Access, Structured Data, Commerce Data, Protocol Support, Security & Trust, and Technical Performance. Weakness in any one can make your business invisible to agents.
• This is not a future problem. The protocols are live, the agents are transacting, and the adoption curve is accelerating. Businesses that prepare now will capture the early wave; those that wait risk being filtered out of agent-driven purchasing decisions entirely.

Where to start

The first step is understanding where you stand. The Zeodyn Scanner™ evaluates your website across all six dimensions of agent commerce readiness and returns a Zeodyn Score™ — a composite metric from 1 to 100 — with specific, actionable recommendations for improvement.

You do not need to implement every protocol on day one. But you do need to know what agents see when they look at your site — and whether that picture is good enough to compete.

Scan your site now — it's free and takes 30 seconds.