What Is Agentic Commerce? A No-Hype Guide for 2026

Agentic commerce is when AI agents research, compare, and sometimes complete purchases on behalf of a shopper, with little or no manual clicking involved. McKinsey estimates the opportunity will reach $3 to $5 trillion in revenue globally by 2030. However, before you rebuild your commerce stack around it, you should know that OpenAI already had to walk back its own flagship agentic checkout feature earlier this year.

What is agentic commerce, exactly?

Agentic commerce is a new form of buying and selling in which an AI agent handles part or all of the shopping journey on a person’s behalf: finding products, comparing prices, and, in some cases, paying for them.

That’s different from a chatbot or recommendation engine. A chatbot answers your questions. A recommendation engine suggests products based on your browsing history. Neither one acts. An agentic system does the shopping – it interprets a goal, plans steps, and executes them across multiple systems without you clicking through each one.

Imagine you tell an agent: “Find me a camping tent under $150, delivered by Friday.” The agent searches retailers, filters by price and delivery date, and either shows you the shortlist or, if you’ve allowed it, completes the purchase. You never open a tab.

Autonomy comes in degrees, though. Most agentic shopping today stops before full autonomy. The agent researches and shortlists; a human still approves the final purchase. Full end-to-end autonomy, where the agent buys without asking, is far rarer and mostly limited to low-stakes, repeat purchases like grocery replenishment.

This distinction matters more than most explainers admit. “Agentic commerce” gets used loosely to describe everything from AI product recommendations to fully autonomous purchasing. When you read a market-size number, check which version they mean.

How big is the opportunity, really?

This is where most articles on this topic get sloppy. They quote a single number, usually the largest, without saying where it comes from.

Here’s what the major research firms actually forecast for 2030:

That’s roughly a 35x gap between the smallest and largest number, for the same term. The difference isn’t forecasting skill. Its scope. McKinsey counts B2B and infrastructure spending; eMarketer only counts checkouts that happen directly inside a chat window.

Narrow the lens to US B2C ecommerce specifically, and the estimates actually converge: most land between 10% and 25% of online sales by 2030. That’s a meaningful shift, but it’s not the overnight replacement of e-commerce that some headlines imply.

What’s already measurable today is the adoption of AI in shopping research, not necessarily autonomous buying. Morgan Stanley found that 23% of Americans made a purchase involving some AI in the past month, with the most activity in groceries and CPG. Bain found 30-45% of US consumers already use generative AI for product research – but about half still won’t let AI complete a purchase without them.

How it actually works: protocols and interaction models

Agentic commerce needs a common language so that agents, merchants, and payment processors can communicate with each other. A handful of protocols have emerged to provide that:

None of these protocols require a business to rebuild its backend from scratch. ACP, for example, is designed to sit on top of your existing commerce and payment systems – if you already process payments with Stripe, enabling agentic payments can be close to a one-line-of-code change.

In practice, these protocols solve narrow, different problems. AP2 handles authorization: a user sets a rule like “spend up to $50 a month on subscription renewals”, and the agent can’t exceed it without asking again. ACP handles the transaction itself: passing order details, a payment token, and delivery info between the agent and merchant, without exposing raw card data to the AI model. MCP isn’t about payments at all – it’s the plumbing that lets an agent pull live inventory or pricing from a merchant’s system in the first place, before any transaction happens.

McKinsey frames the actual buying flow as one of three interaction models: a person’s agent talking directly to a merchant’s site, one agent negotiating with another agent on the merchant’s side, or a broker sitting between the two and mediating the transaction. Which model wins for a given category is still an open question – and depends heavily on which protocol merchants and platforms actually adopt at scale.

Reality check: what’s actually live in 2026

Here’s the part most explainers skip: agentic checkout has already had a very public stumble.

In September 2025, OpenAI launched Instant Checkout, enabling ChatGPT users to buy directly from Etsy sellers, with “over a million” Shopify merchants promised to follow suit. It was framed as the moment AI shopping went mainstream.

Six months later, reality looked different. By March 2026, roughly 30 Shopify merchants had actually gone live with Instant Checkout, according to Forrester analyst Emily Pfeiffer. OpenAI pulled back from native in-chat checkout entirely, shifting to an app-based model with purchases redirected to the merchant’s own site. Onboarding merchants turned out to be far harder than announced, and the checkout experience itself was reportedly error-prone.

This isn’t a story about agentic commerce failing. It’s a story about the gap between announcement and infrastructure. Payments, fraud checks, tax collection, and returns are hard problems that don’t disappear just because an AI model is involved. And not every retailer wants agents in the mix at all: Amazon has not only actively blocked ChatGPT from accessing its site, the e-commerce giant sued Perplexity over its Comet browser-scraping product-pages-for-autonomous-purchases product.

What is working and growing fast is AI-driven traffic and research, not necessarily AI-completed checkouts. Adobe Analytics recorded a 693% year-over-year jump in traffic to US retail sites from generative AI sources during the 2025 holiday season. Grocery and CPG replenishment, where the decision is low-risk and repetitive, remains the strongest real-world use case for actual autonomous purchasing today.

What this means if you’re building a mobile or e-commerce app

Whether or not full autonomous checkout takes off on your platform, one thing is already worth doing: making your product data legible to agents, just as you’d optimize it for search engines.

That means structured, accurate, machine-readable product feeds: price, availability, variants, delivery windows, return policy. Agents don’t parse a nicely designed product page the way a human does. They need APIs and clean data, not visual design.

Concretely, this usually means schema.org Product markup on every product page (price, availability, and SKU as structured fields, not just visible text), plus a real-time feed or API endpoint an agent can query directly instead of scraping your storefront. If your product data currently lives only in your CMS templates, with price and stock rendered as HTML text, an agent has to guess at what it’s reading. A clean feed removes the guesswork.

This is also where mobile commerce architecture decisions start to matter for reasons beyond user experience. An app built on a headless, API-first backend is naturally closer to “agent-ready” than one where product data, pricing, and inventory logic are tangled into a monolithic frontend. If you’ve followed our mobile commerce guide, the same architectural principles that support fast checkout and omnichannel sync also make it easier to expose clean data to agents later, without a rebuild.

None of this requires picking a winning protocol today. It requires not locking yourself into a rigid, tightly coupled backend that will need to be torn apart the moment agent traffic becomes material to your sales.

Risks and trust gaps

The biggest obstacle to agentic commerce isn’t technical. It’s trust.

IBM’s 2026 research found 83% of consumers share concerns about privacy or data misuse when it comes to AI handling their purchases. Handing an agent your payment details and delivery address requires a level of confidence most people aren’t at yet, especially for anything beyond low-value, repeat purchases.

There’s also a builder-side risk worth knowing about. Gartner predicts more than 40% of agentic AI projects will be canceled by the end of 2027, citing unclear business value, escalating costs, and inadequate risk controls. Gartner also flags “agent washing”: vendors rebranding ordinary chatbots or automation tools as “agentic” without real autonomous capability behind them. It’s worth remembering when a vendor pitches you an agentic commerce solution.

For merchants, the practical risks are fraud (can you tell a legitimate agent from a scraper?) and accountability (who’s liable when an agent buys the wrong item?). Regulation is also still catching up: EMARKETER expects the EU’s regulatory environment to slow investment in agentic commerce and AI-generated ads through 2026, well ahead of any settled rulebook for autonomous purchasing specifically.

None of this means waiting and doing nothing. It means treat agentic commerce as an infrastructure investment with a realistic timeline, not a feature you bolt on this quarter.

Key takeaways and next steps

This week: audit whether your product data (pricing, stock, delivery windows, and return policy) is exposed through a clean API, or buried in frontend templates only humans can read.

This quarter: decide whether your business benefits more from being discoverable inside third-party agents (such as ChatGPT, Perplexity, and Gemini) or from building your own on-site shopping assistant. They require different investments.

Looking further out: expect the protocol picture to keep shifting through 2026 and 2027. Building on API-first, headless architecture protects you either way, because you’re not locked into whichever protocol happens to be dominant this year.