A-Commerce and the Illusion of Simplicity

Why agentic commerce isn’t ready for prime time – and what we need to fix first

There’s a growing wave of excitement in tech circles about Agentic Commerce – the idea that AI-powered agents will soon handle our shopping for us. From Fast Company to VC posts on LinkedIn, everyone seems to be leaning into this future. It sounds magical: “ChatGPT, buy me the perfect dress for an event next weekend” – and voilà, it’s done.

But for those of us who have been in the commerce trenches before, this isn’t a new idea. In fact, we had another name for it: Universal Commerce.

In 2021 and even earlier, many of us, Carted included, tried solving programmatic, multi-merchant checkout. We built systems that could place orders anywhere on the internet without needing merchant-by-merchant integrations or affiliate feeds. It was a technical feat, and we shipped it across media sites, enabling instant checkout experiences directly from product links. We thought we’d solved a major friction in ecommerce. But what we learned was that checkout isn’t the problem. The real friction lives upstream – discovery, comparison, validation, preference, trust.

Consumers don’t always “see something and buy it.”

They check whether it’s in their size, from a store they trust, at a price they’re willing to pay. They compare it with some other options. They want to use a specific credit card to earn points or a Buy-Now-Pay-Later option that fits their current budget. They log in for loyalty rewards, look for coupon codes in their inbox, and often send links to friends for a second opinion.

So even with a billion products in the database and a perfectly functioning checkout, people didn’t convert. It wasn’t enough. We shifted priorities to get closer to that intent layer – to understand what people actually need before they ever hit “checkout.”

Now we’re seeing Agentic Commerce re-emerge, only this time the interface is a chatbot, powered by AI, rather than embedded in the source media articles. And while the tech may be slightly different, the same fundamental limitations are still holding it back.

Frankly, I don’t think agentic commerce is going to take off any time soon – at least not in the way many think it will over the next two years. There are still hard problems that need to be solved that aren’t unlocked by LLMs.

For starters, most consumers don’t actually want to chat to shop. There are some cases where it makes sense – reordering household staples, for example – but shopping is rarely a straight-line task that you complete in one sitting. It’s emotional, social, and often playful. On any given Saturday, if you walk down Fillmore Street in San Francisco, you’ll see friends, mothers and daughters browsing boutiques together. You’ll overhear conversations about what to wear to a wedding, what to pack for Palm Springs. That joy, that sense of exploration and discovery, isn’t something a chatbot can replicate. The interface will matter.

Even if consumers are ready, the merchant side of the equation isn’t. For agentic commerce to work, merchants need high-quality, structured data – real-time inventory, product variants, pricing, shipping info, return policies. But most don’t have that structured data themselves. Even major retailers fall short. This week, I was in Australia and tried to buy sneakers online. They wouldn’t arrive in time, so I wanted to pick them up in-store. I couldn’t figure out which location had stock because the retailer’s website couldn’t tell me. I called three stores. No one answered. I didn’t buy the shoes. It would be great if an agent could call for me and check but until the store has an agent that has access to that information, it won’t happen.

At ShopTalk this year, the GM of Google Shopping called out this exact issue – urging merchants to improve their product feeds. Without accurate, structured data, even the best AI agents are flying blind. And this isn’t a new problem. Instagram Shopping struggled with the same thing. Merchants had to opt in, and adoption was inconsistent. The experience became fragmented and frustrating. Shoppers abandoned it.

On top of that, shopping preferences are deeply nuanced. Since moving to the U.S., I’ve had to get multiple credit cards to build a credit score. Which card I use at checkout depends on rewards, balances, how I’ve allocated funds that month, and whether I want to route the purchase through a BNPL provider. Can an agent eventually learn that? Maybe. But it requires knowing me better than I know myself – and being integrated with my financial life in a way that most people simply aren’t comfortable with today. And that’s just payment preferences, let alone my style tastes!

The excitement around agentic commerce is understandable. The future where an AI can read your mind, access your shopping history, understand your taste, budget, urgency, and social dynamics – and deliver the perfect purchase experience – sounds amazing. I’ve imagined it too. But we don’t live in that world yet. We live in a world of walled gardens, privacy regulations, incomplete data, unstructured merchant feeds, and unpredictable human behaviour.

What we need is a renewed focus on the foundational layers: structured product data, real-time availability, clean variant-level detail, and systems that account for how people actually shop – not just how they transact. Before we can run, we need to crawl. Checkout is only the final click. The real complexity lives in everything that comes before.

So the next time you see a viral post about agentic commerce and feel the urge to comment “This is amazing!”, I challenge you to try solving your next real shopping problem using one of these tools. Not a hypothetical task. A real need. Something you were actually planning to buy. You’ll likely see how many invisible, unresolved steps still exist.

Those are the layers we need to build, and problems we need to solve. That’s where the real opportunity is.