Retailers – there is a path to agentic commerce, but no shortcut

Sometimes, trends in technology move so fast that we all feel behind. AI – and AI search especially – is a defining example.

ChatGPT – the most prominent of a host of AI platforms – famously acquired a million users within five days of launching in 2022, and 100m in about two months. As of mid-2025, it was used by roughly 10% of the earth’s adult population. Search will clearly never be the same again.

But while people can leap on a trend in seconds, brands, advertisers and infrastructure builders have taken longer to move. And if a large segment of the world’s population has a new way to search, then advertisers who depend on search traffic need to figure out how to follow their audience – and fast.

So far, AI search is a space that largely exists outside the scope of marketing – but not for long. ChatGPT is, of course, bringing conversational ad units, and Particular Audience has made moves with Retail-MCP.com to build the foundations for ‘Functional’ Ad Units, but there is a bigger opportunity here than new ad formats alone.

At PA, we’ve long believed commerce eats advertising, as discovery and conversion converge everywhere an ad could have appeared. Ultimately, agentic AI offers the opportunity to collapse commerce and advertising into a single system, where ads are no longer distinct from discovery, and conversion is no longer a downstream metric, but becomes the ad product itself.

Agentic commerce needs ads

Of course, there are technological complexities here. OpenAI adopted the Model Context Protocol (MCP) as the standard for building integrations (now called ‘Apps’) for ChatGPT. MCP is an open-source standard that enables AI models to connect seamlessly with external tools and data.

Up until now, however, apps have had a distribution problem, in that filtering all apps on the basis of real-time chat hasn’t been economically viable at scale. Advertising changes the economics of this, introducing funding that makes selective, intent-driven invocation of functionality viable.

In simple terms, we want AI to be better integrated and more functionally capable, but it would cost too much to call functionality from all the brands in the world every time intent is signalled. Ad revenue can, however, foot the bill.

When that happens, with retailers building functional apps of their own, ads in AI search become functional shopping experiences. And with AI assistants as the new storefront, the entire purchase journey can play out in one place. Brand chat, the main aim of which was to offer some information and possibly a link to purchase, now provides the entire outcome you were trying to achieve, from research to conversion.

In this vision, ads stop being messages and become real-time interfaces; checkout becomes a tool, not a redirect; loyalty, order status and support are all there too. Even negotiation becomes feasible.

The technical requirements of agentic commerce

Anyone can imagine the seamless glory of agentic commerce. But to stand a chance of getting to that point, retailers and Retail Media Networks (RMNs) urgently need to ask themselves a few questions:

1. Do we have an MCP architecture for agentic interoperability with our systems?
2. Does this architecture support existing merchandising and monetisation rules to survive and thrive in AI interfaces?
3. What differentiated rich experiences optimise outcomes when interfaces are interactive?

While the ability to bring transactional capabilities to AI chat is essential, it is also just one function; MCP architectures need to govern decisioning and enable all ways customers might want to interact with a brand, not just facilitating payment.

For brands and retailers, using MCP to solve the specific engineering challenges of agentic commerce requires a dual-layer stack. This consists of a Transaction Layer – which standardises the handshake between the AI agent and the checkout (or other read/write system) – and an Intelligence Layer, which not only ingests your inventory but infers meaning, vectorises it, and enables business objectives to compete within the ranking cocktail that ultimately governs relevance.

In other words, when an agent requests ‘waterproof gear for high-humidity climates’, it is able to retrieve semantically accurate SKUs, rather than zero-result keyword failures or obscure token matched results, in a low latency response.

Owning the handshake

The MCP handshake – the process that establishes secure, standardised connections between AI models and external data sources or tools – is now being adopted widely across the AI industry as a standard for AI-tool communication. It is vital that retailers, not platforms, own their slice of this layer – their profitability depends on it.

The good news is that if MCP is a free-to-use open source standard, and retailers and brands only need to build their MCP tooling once. They can then traffic the intelligence into ChatGPT, Claude, and others, solving the fragmentation problem in the process.

On the other side of these layers, the opportunity is enormous – but there is no shortcut to get to it. Which is why, for brands hoping to capitalise on the imminent wave of agentic commerce, Retail-MCP orchestration is the most scalable investment they can make right now. 

Also published in: Modern Retail