Morning All,

I've said for a while that 2026 is the year of the AI agent. Even if you don't agree, at the very least it's the year that AI agents went mainstream. We've known for a while that AI is moving with a Thanos like inevitability when it comes to taking over our job tasks. But what about the things we (at least some of us) enjoy doing? The world economy is built on a foundation of people working, earning money, and then using that money to buy stuff we don't need to impress people we don't like. What does life look like when even that is done by AI systems? There's an interesting discussion to be had about what does life look like for consumers in that scenario, but today we're focusing on what does that reality look like for retailers.

The 6 layers of AI

If you remember, in the edition from 2 weeks ago I mentioned the 6 layers of the AI economy. In a previous life I did buying and merchandise planning for Nike, Ralph Lauren, Kurt Geiger amongst others. DTC e-commerce and brick & mortar retail. So when I think of AI in retail, that's my frame of reference...at least for today. How is AI affecting the transactional relationship between brands, retailers, and their customers. There are many different discussions to be had about how AI is affecting the operational side of running a retail business. Supply chain, S&OP, demand forecasting, PLM, pricing and inventory management, all of which we're not going to cover today. (But definitely will at a later date). And at this point I should be clear that when I say AI, I specifically mean Gen AI, LLM's etc and not traditional machine learning like analytics and computer vision etc. Alright? good.

*The Agent & Tools Layer is all about IoT (Internet of Things) and it's main use case is supply chain so we're not going to cover it today. Let's dive in to the rest...

When thinking about each of the layers, they can be split between direct control or 3rd party partnerships.

The Infrastructure Layer (3rd Party Partnership)

As a by product of e-commerce, modern retail is already mainly cloud based. Through AWS, GCP or Azure, retailers already have access to the scale of compute they would need. Consequently, investing in their own data centres and compute, doesn't seem like the most efficient allocation of resources for nearly all retailers. There are obviously a few notable exceptions...Amazon, Apple, Walmart for example. However, the industry standard practice is to rent the infrastructure layer from a cloud services provider.

The Data Layer (Direct Control)

The most critical layer. This is the layer that creates the value to be extracted from every other layer. Most retailers do not have the unified data needed to deploy a vertical AI solution that can handle tasks from end to end (Layer 6). Most retailers don't have multi channel sales, inventory, and traffic, as well as customer and supply chain information, all in one centralised system. That's just for starters. Making this data layer the most effective it could be is the key step. The nitty gritty of what this looks like is too technical for this newsletter, although I will find a forum to write about it. Essentially, get this wrong and an AI system will only amplify and accelerate the problems you've already created.

The Model Layer (Direct Control)

An underrated opportunity. I'm a big believer in the power of open source AI. Fine tuning a smaller foundation model on your own company data will yield much better results vs simply giving ChatGPT access to your product catalogue. Don't take my word for it, ask Walmart who cancelled a multi-billion dollar deal with OpenAI in favour of working with their own internal models. "We learned that our customers want consistency across every touchpoint," and simply connecting to ChatGPT's instant checkout feature lead to poor order accuracy and a significant drop in conversion rates. Google, Meta, Mistral, and a whole range of Chinese labs have released open source models that are extremely powerful, and almost as good as OpenAI and Anthropic models...and they will only get better. A huge opportunity for any retailer who wants to seize it.

The App Layer (Direct Control)

This might be the layer at which every retailer does most of their work. Whether it's adding conversational AI features to existing sites, or using AI generated content in marketing, customer service etc. Sephora uses AI-powered virtual assistants to provide personalised beauty consultations based on skin tone and purchase history, enhancing customer engagement. DSW’s AI chat agent handling complex returns and exchanges, saving $1.5 million annually.

What does agentic commerce mean for retail and retailers?

Successful agentic commerce will depend on executing well at every level we've just been through. Sounds like a lot work, because it is. But the opportunity and the rewards are huge. For a select few retailers and brands it'll make sense to have their own autonomous AI personal shopping agent. Department stores & supermarkets come to mind. For most, my prediction is their relationship with agentic commerce will be similar to their current relationship with Google Search...mainly about product discoverability.

The scale of what's coming is hard to ignore. Morgan Stanley predicts that between 10% and 20% of all US e-commerce spend could be driven by agentic AI shoppers by 2030. Deloitte goes further, citing analysts who believe 25% of global e-commerce sales will be enabled by AI agents by that same year. Early market data shows 25-40% of users in developed markets already rely on AI tools for discovery and comparison.

Carrefour has already integrated its grocery catalogue directly into a ChatGPT interface, allowing AI agents to check real-time availability, build baskets, and hand off to checkout. We'll see if they find joy where Walmart only found frustration. Either way, what is abundantly clear is that retailers not set up for this kind of discoverability are effectively invisible to an increasingly large slice of the market. Platforms and brands now need to optimise for algorithm attention, not just human attention. Buying influence is shifting from product pages and browsing funnels to the signals agents consume: structured metadata, pricing, availability, delivery guarantees, and trust signals.

• Structured metadata: canonical product identifiers, machine-readable categories, attributes, and variant normalisation
• Real-time signals: pricing, inventory, delivery windows, and return policies
• Reputation and trust data: seller ratings, provenance, and dispute-resolution history
• Integration endpoints: programmatic checkout, consented payment credentials, and webhook-based updates

What's stopping retailers getting there today?

The legal and contractual headaches are significant. Under English law, AI systems have no legal personality, meaning they can't actually be parties to a contract. That puts the burden on retailers to verify the scope of authority an AI agent has been granted, and to update their terms of purchase accordingly. Right now, most retailers' website terms were written long before agentic transactions existed...and many accidentally prohibit them.

Consumer protection obligations don't disappear just because a human isn't clicking the buttons. UK law requires that consumers actually understand key information like cancellation rights and returns policies at the point of purchase. Delivering that information to an AI agent almost certainly doesn't count. Retailers need to rethink how they confirm a real human has understood and consented to a transaction.

Card scheme rules assume a cardholder is directly involved in a purchase. Agent-initiated transactions sit awkwardly outside that framework, exposing retailers to elevated chargeback rates. Worse, traditional fraud detection systems may flag legitimate AI agent activity as bot behaviour, blocking valid sales rather than protecting against bad ones.

What could the future look like?

China's AI Agents Are Already Running Errands For People

While most of us think of AI as a tool that helps us make decisions, China is already moving past that paradigm. Platforms there are deploying AI agents that don't just suggest what to buy...they actually complete the purchase, handle the logistics, and check back in with you only when something unexpected happens. This is The Vertical Solution Layer - Level 6 of the AI pyramid, deployed at scale.

When Meituan launched its Xiaomei AI agent in late 2025, executives internally described it not as a chatbot but as an "orchestrator plus execution agent." A user can say, "Order my usual lunch, but deliver it 20 minutes later today," and the agent interprets the intent, applies stored preferences, and completes the transaction, often with zero screen interaction. This creates a meaningfully different relationship between a person and a platform and Meituan isn't alone. Alibaba, Ant Group, and ByteDance are all testing similar systems. China infrastructure and the maturity of the consumer makes it an ideal proving ground. Widespread digital payments, dense logistics networks, and super-app platforms that already stitch together dozens of services create the conditions for agentic commerce to actually work. You can't delegate a task to an AI if the underlying pipes aren't solid.

The "super-app" model matters more than most Western readers will appreciate. Apps like Meituan combine services similar to JustEat, Yelp, and Groupon into a single platform. When your payment system, your merchant network, and your delivery logistics all live under one roof, an AI agent can move across all of them seamlessly. In markets where those layers are fragmented across different companies, that kind of fluid delegation is far harder to pull off.

Remember I mentioned earlier in this piece that government regulation is an open questions...well in China that's less of a problem. China's approach allows experimentation before formal rules emerge. That creates a window for platforms to test, iterate, and learn quickly, without waiting for regulatory clarity that may never come. Whether that's a feature or a risk depends on your perspective, but it's undeniably a competitive advantage.

Agentic AI is primarily reshaping retail from the demand side, not the supply side...at least not yet in the Western hemisphere. The retailers who win will be those who make their products easy for AI systems to find, understand, and buy, while simultaneously tightening their legal, fraud, and compliance frameworks before regulators force their hand. When the supply side shift lands in Western markets, and it will, the companies that'll win won't just be the ones with the best models. They'll be the ones that control enough of the AI stack for an agent to actually do something useful from end to end.