(written jointly with Greg Lewis)
Imagine a future where buying things is as simple as having a conversation with your phone or computer. You describe what you're looking for, and an AI agent guides you all the way from vague need to checkout. Something hot for a Brat summer? No problem. The agent knows what you like - it has your past purchase data. It knows what people like you like. And it knows what a Brat summer is without you having to explain it.
Even better, like a good friend, it’ll ask smart questions to help you find the perfect outfit. “What’s your price point?”, “Is this for the beach or a party?”. Once you've made your choice, the agent handles the purchase seamlessly - no irritating checkout process, no need to remember logins or shipping addresses.
This vision may seem futuristic, but the building blocks already exist. Generative AI agents like Chat-GPT or Claude already understand context, can ask thoughtful questions, and can voice chat. Experimental versions of Anthropic’s product can already use computers just as a human would. What’s missing is the “pipes” that connect these agents to merchants and their product catalogs, as well as personalization of the agents based on your own preferences and the current cultural zeitgeist. But with this in place, finding what you truly want will become easier than ever.
The Opportunity and The Key Players
These agents will insert themselves in between the customer and the most profitable business model in history, performance advertising. As a result, the market opportunity is huge. We envisage two main categories of players: “vertical agents”, who are integrated into a vertical stack owned by an existing e-commerce platform, and “horizontal agents” who work across platforms to find customers the best deals.
Vertical agents will be the first to market. Amazon, with its vast product catalog and customer data, is particularly well-positioned to deliver a conversational shopping interface that reduces friction for buyers. In fact, it is already experimenting with a shopping bot called Rufus. Vertical agents will excel at guiding customers through a merchant's offerings, but their motives may be suspect. For example, Amazon may direct its agent to select Amazon-branded or higher margin products, as some believe Amazon currently does with its search results.
Horizontal agents, on the other hand, will search across the web, interacting with multiple merchants to find the best options. These will be especially valuable in categories where no single merchant dominates, such as clothing, healthcare, or travel. While booking sites may have a wide selection of hotels, they often lack other components like flights or tours. Horizontal agents can piece together the perfect trip from various sources. We see foundational model developers such as OpenAI, Anthropic, Meta, and Google as potential players in horizontal shopping agents. Meta and Google have especially large incentives to build these, since AI agents threaten their business model.
Monetization Strategies
How will these AI shopping agents make money? Advertising will remain a key revenue stream - bots will make paid recommendations. But the nature of ads will evolve to be considerably more sophisticated.
Rather than traditional ads with uninformative text, we'll see a rise in ad copy tailored to the conversational context and user demographics. The agents themselves will design these ads on the fly, figuring out what product features might best appeal to the customer, referencing a content library supplied by the advertisers. Maybe they’ll hype a celebrity tie-in, or emphasize a particular design feature, or a strong product review. In this way, these agents will displace both shopping and copywriting.
Advertisers might even hand over sales strategy to the agents, allowing them for example to offer special discounts of up to $500K over the course of a month to users of the agent, in order to increase conversion. So if an agent senses that a customer is wavering, it might present a time-limited discount to push them over the edge and make the sale, just like an excellent salesperson.
But unlike great copy-writers and salespeople, who typically work for a single company, these agents have many clients. And so they have the opportunity to build on one successful transaction by suggesting complementary products - the perfect accessories for that dress, for example. They might even bundle these products together, offering a deal for buying both, to the benefit of both of the companies they represent.
There is still the natural tension between organic and paid content. Agents will be tempted to favor the content they are paid to promote. But the level of personalization will generally weaken these incentives, as paid content will become jarringly blatant if totally out-of-context. Big advertisers will be unable to “dominate the airwaves” in places where their products just don’t fit.
What should businesses do differently?
In this new landscape, external press, and reviews will grow in importance since agents will look for credible information sources to pick products to recommend. By contrast, the value of ad copy in performance marketing will diminish since agents will essentially "write their own copy" based on product data.
Perhaps surprisingly, branding will remain critical even with AI agents. A key part of the brat summer outfit for some will be the brand that their friends will recognize or ask about. More generally, consumers care about brands as a way to signal to others and themselves about their identity. Shopping agents will need to account for the brand when making recommendations, and will use an immense amount of information online (Reddit, YouTube, etc…) to learn about the brand.
Another strategy in response to AI shopping agents is to expand the product line and to allow for more customization. AI agents will be able to consider all the products. Large catalogs that once overwhelmed customers will become an asset, helping bots find the perfect item. At the same time, obfuscatory marketing strategies will become less effective, as bots will be unlikely to be tricked in these ways. For example, well-designed bots will be aware of add-ons, hidden fees, and other dark patterns.
What do the above three strategies have in common? They better align customers and the products they purchase, which is a good thing for both customers and quality businesses.
If you’re working on AI agents and are interested in collaborating, please reach out!
Note: Greg and I wrote this in order to submit to Harvard Business Review, but it was deemed too ‘theoretical’ so you get to enjoy it here!