Agentic Commerce

E-commerce has evolved in waves: Storefront, mobile, social. The next one is agentic commerce, where AI agents don’t just help people shop, but shop for them.

What Is Agentic Commerce?

Agentic commerce describes a model where autonomous AI agents, powered by large language models and tool-use capabilities, act on behalf of consumers to research, compare, negotiate, and even purchase products and services. Instead of a human browsing a website, clicking through filters, and adding items to a cart, an AI agent handles the entire journey.

Think of it this way: today, a customer might ask ChatGPT or a similar assistant, “What’s the best running shoe for flat feet under 150 €?” The AI responds with recommendations. That’s assistive AI. In agentic commerce, the customer says, “Buy me the best running shoe for flat feet under 150 €, and have it here by Friday.” The agent researches options, compares prices and availability, and presents a recommendation with a one-click checkout. The human still confirms before money is spent, but the agent did all the work. That’s a fundamentally different interaction.

But there’s a third scenario, and arguably the most disruptive one. Imagine someone sends their AI assistant a photo of a broken fence and asks, “How do I fix this?” The agent walks them through the repair steps and then, unprompted, says: “You’ll need a specific bracket, exterior wood screws, and weather-resistant gloves. Here are options that match, available for delivery by tomorrow.”

The user never intended to shop. There was no purchase intent at the start of that conversation. The agent identified a need during an unrelated interaction and surfaced relevant products in context. Commerce didn’t begin at a storefront or a search bar. It emerged from a conversation about home repair.

This is what makes agentic commerce fundamentally different from everything that came before. It’s not just a faster checkout. It’s commerce that can happen anywhere, inside any conversation, whenever an AI agent recognizes a need, it can fulfill.

And in all of these scenarios, the customer never visits your website. They never see your brand colors, your carefully designed landing page, or your promotional banner. The AI agent interacts with your data and APIs, not your frontend.

When an agent can compare prices across dozens of merchants in milliseconds, the game shifts fast and the major AI companies are actively accelerating that shift.

The Incentives of AI Companies, and What That Means for You

The major AI companies are deeply incentivized to make agentic commerce happen.

OpenAI and Google are building ad-funded commerce channels. OpenAI projects that up to 20% of its revenue could come from advertising and sales commissions by 2029, potentially a $25 billion business. Google, meanwhile, has an entire search-ads empire to protect. If consumers shift from Google Search to AI assistants for product discovery, Google needs commerce deeply integrated into Gemini to keep that revenue stream alive. For both, agentic commerce is a natural extension of their advertising business: sponsored placements, premium visibility, and transaction fees.

Anthropic is the outlier. Squarely B2B and explicitly anti-ad, their incentive is to become indispensable infrastructure for enterprises rather than take a cut of consumer transactions. For merchants, that means fewer intermediary fees and more control over the customer relationship, in theory. The catch: there’s no simple “pay to be visible” lever. With OpenAI or Google, you can buy your way in via ads or sponsored placements. With Anthropic, the path to being surfaced through Claude-powered agents is less direct and still largely undefined.

Two things stand out. First: you are not the customer — until you become one. These platforms are building for consumers. If your store is easy for agents to interact with, you benefit organically. If not, you get skipped. But as sponsored placements arrive, you can pay to be surfaced, much like Google Ads or Meta today. Either way, the platform wins.

Second, and more importantly: early movers have leverage, for now. These platforms need merchants to get their ecosystems off the ground. Without product feeds, checkout integrations, and fulfillment partners, there’s nothing for agents to sell. That gives early adopters a rare window to negotiate better terms, shape how protocols evolve, and secure organic visibility before paid placements take over. Once merchant participation becomes table stakes, that leverage disappears.

The Protocol Landscape: MCP, ACP, UCP, and the Walled Gardens

If you’re wondering how all of this actually works on a technical level, the answer is: through a growing set of protocols that define how AI agents talk to merchants. Three open protocols and a set of proprietary platforms are shaping the field right now.

The Model Context Protocol (MCP), originally developed by Anthropic, is the most general-purpose option. It’s not commerce-specific but rather a universal interface that lets AI applications connect to external tools and data sources. For commerce, that means building an MCP server that exposes your product catalog, inventory, cart, and checkout flows. Shopify already offers official MCP servers for its merchants. The big advantage: MCP is platform-agnostic, open, and free to use with no vendor lock-in. The downside is that it wasn’t designed with commerce in mind, so it lacks native payment handling, and the incentives for AI assistants to actively use your MCP server aren’t always clear.

OpenAI’s Agentic Commerce Protocol (ACP), launched in September 2025, is purpose-built for end-to-end purchasing: product discovery, checkout, and fulfillment, all within ChatGPT. It’s open source under Apache 2.0 and co-developed with Stripe. Merchants provide product feeds and checkout APIs, and Stripe handles payment processing. The protocol powers ChatGPT’s “Instant Checkout” feature, which now reaches 800 million weekly active users. Currently, results are organic and not sponsored, but that will likely change. The limitations: for now, it’s restricted to approved partners and only works within ChatGPT, and transaction fees apply.

Google’s Universal Commerce Protocol (UCP), announced in January 2026, takes the broadest approach. Co-developed with Shopify, Target, Walmart, Etsy, Wayfair, and endorsed by over 20 partners including Visa, Mastercard, and Salesforce, UCP is designed to work across Google Search, Gemini, and beyond. It requires Google Merchant Center integration and supports checkout, identity linking, and order management. It’s compatible with MCP and other protocols. The reach is massive, but it’s very new, initially US-focused, and given Google’s business model, expect visibility to eventually be tied to ad spend.

Then there are the proprietary platforms. Amazon’s “Buy for Me” feature lets its AI agent purchase products from third-party websites on behalf of shoppers, even without the merchant’s explicit participation. It sparked significant retailer backlash in early 2026 when merchants discovered their products were being listed and sold without their knowledge. Microsoft offers Copilot integrations, and independent players like Perplexity are building AI shopping agents that aggregate across merchants. These platforms offer access to large user bases and handle the UX, but the trade-off is real: you give up customer relationships, operate inside walled gardens, and in Amazon’s case, you may not even have a say in how your products are presented.

Here’s how the options compare at a glance:

The strategic takeaway: don’t bet on a single protocol. MCP gives you vendor independence, ACP gives you access to ChatGPT’s massive user base, UCP plugs you into Google’s ecosystem, and proprietary platforms offer reach at the cost of control. A multi-protocol approach, combined with a strong owned storefront (the one channel you fully control), is the most resilient strategy.

It’s More Accessible Than You Think

Here’s the good news: getting started with agentic commerce readiness is not a massive technical lift for most modern e-commerce operations.

If you’re running on Shopify, commercetools, or any commerce stack, you likely already have what you need. And if you’re running a custom-built e-commerce solution, the effort is absolutely manageable: the protocols are well-documented, and the integration points are standard REST or GraphQL APIs. Either way, the work is less about building new systems and more about exposing and optimizing what’s already there:

1. Ensure your product data is rich, accurate, and structured. Agents respond to structured information, not visual merchandising. Incomplete descriptions, missing attributes, and inconsistent formatting are your biggest enemies.

2. Make your APIs agent friendly. This means clear documentation, reliable uptime, sensible rate limits, and responses that are easy for an LLM-based system to interpret.

3. Implement or adopt emerging agent protocols. Keep an eye on standards for agent-to-merchant communication. Early frameworks are already appearing and adopting them puts you ahead of the curve.

4. Think about authentication and trust. Agents acting on behalf of consumers need secure, scoped access to place orders. Consider how you handle delegated authorization.

For most businesses, the technical integration is a matter of weeks, not months. The harder work is strategic: rethinking how you communicate value when your audience is an algorithm, not a human.

A note for European merchants

As of early 2026, all three major agentic commerce initiatives, ACP, UCP, and Amazon’s Buy for Me, are live in the US only. None have announced firm European launch dates. OpenAI has signaled “international expansion” within 2026, and German trade press expects a rollout in the coming months, but nothing is confirmed. Google faces additional EU regulatory hurdles that could delay UCP’s arrival. For European e-commerce companies, this creates a paradox: the window to prepare is open precisely because the features haven’t arrived yet. The merchants who use this lead time to get their data, feeds, and API readiness in order will have a head start once these platforms go live in the EU.

The Bottom Line

Agentic commerce is an additional distribution layer, not a replacement for what you’ve built. The merchants who treat it that way early, while the leverage is still on their side, are the ones who’ll be in the room when the rules get written.