Google go all in on Agentic Commerce

Last week at the National Retail Federation conference, Google dropped something big. Not just another feature update or incremental improvement, but a proper attempt to solve one of the messiest problems in agentic commerce: protocol fragmentation.

They call it the Universal Commerce Protocol, and if it works as intended, this could be the moment when AI-powered shopping moves from experimental to something retailers can really build on.

The protocol problem

Here's the issue currently: we've seen multiple companies create their own standards for AI agents to facilitate commerce. OpenAI has their Agent Payments Protocol (AP2). There's Agent2Agent (A2A) for inter-agent communication. Anthropic is pushing Model Context Protocol (MCP). Each one serves a purpose, but it can be confusing.

The result? Retailers face too many options. To support AI shopping across different platforms, they'd need to integrate with each protocol separately. It's inefficient, expensive, and exactly the kind of friction that kills adoption before it starts.

Google's solution is UCP, a protocol designed to sit above the existing standards and create interoperability. Think of it as a translation layer that lets different agent protocols work together without requiring retailers to rebuild everything for each one.

UCP has backing from Shopify, Etsy, Wayfair, Target, Walmart, plus payment processors like Stripe, PayPal, Adyen, and the major card networks.

What this means for retailers

UCP covers the full shopping journey: product discovery, cart management, checkout, order tracking, and post-purchase support. Google is also launching 'Business Agent', which lets shoppers chat with brands directly on Search, functioning like a virtual sales assistant trained on the retailer's catalogue and voice.

They're also expanding the Merchant Centre with dozens of new data attributes designed for conversational commerce. This is critical, because current product feeds were built for keyword search, not for AI agents that need to understand compatibility, substitutes, common questions, and use cases. If you want your products to surface well in AI-driven discovery, you'll need to feed these systems richer context.

There's also Direct Offers, a new ad format that lets retailers present exclusive discounts directly in AI Mode when shoppers are researching products. It's Google's first proper 'agentic ad' unit, and while the initial focus is on discounts, they're planning to expand to bundles and free shipping.

The checkout

Google is also upgrading their 'Buy for Me' feature, which had some limitations in its first iteration. The new checkout will appear on eligible product listings in AI Mode and the Gemini app, supporting Google Pay initially with PayPal coming soon.

Retailers remain the merchant of record, which matters for brand control and customer relationships. And unlike the first version, which had a 'percent of a percent of a percent' reach (you had to be in AI Mode, on a shopping query, with a participating merchant), this should have broader reach.

Why this matters now

We're seeing multiple forces converge. AI agents are getting better at complex reasoning. Multimodal models can understand images and video. Tools like ChatGPT and Gemini are where people are already starting their research. Adobe reported that AI-driven traffic to retailer sites grew 693% during the 2025 holiday season.

The question isn't whether AI will change how people shop. It's whether the underlying infrastructure will be open enough to prevent another walled garden situation where a few platforms control access to customers.

Google positioning UCP as an open standard is important. The fact that it's designed to work with existing protocols (A2A, AP2, MCP) rather than replace them entirely suggests they're serious about interoperability.

The challenges ahead

We are still early with Agentic Commerce, and UCP in particular. The expanded product feed attributes are rolling out to a small group first. Direct Offers is a pilot. Business Agent just launched. None are available in Australia yet.

The real test will be adoption. Will mid-sized retailers invest in richer product data? Will payment processors integrate smoothly? One thing I have no doubt of is that consumers will complete purchases through AI interfaces at scale, I believe once Agentic Commerce is widely available, the sales will flow through.

It will be really interesting to see if OpenAI (with ChatGPT) and other platforms will adopt UCP or push their own standards. Competition in this space is fierce, and no one wants to cede control of the shopping layer.

For Australian retailers watching this unfold, the message is clear: start thinking about how your product data works in a conversational context. The feed specifications built for keyword search won't cut it when AI agents are doing the discovery.

This is moving fast. Like most things in AI right now, what seems experimental today might be table stakes in 12 months.