The boring layers win.
Nobody thinks about TCP/IP, but the internet is unthinkable without it. Nobody thinks about the shipping container, and it quietly reorganized world trade. Nobody thinks about ACH when their paycheck lands. The most valuable infrastructure tends to be the kind you stop seeing — the layer everything else is allowed to take for granted.
Fintech had its version. For a long time, the hardest part of building a financial app wasn’t the app. It was the plumbing — every developer writing brittle, one-off integrations to pull data out of banks that never agreed on a format. Plaid standardized that messy seam into a single clean interface. It became the layer underneath Venmo, Robinhood, Coinbase, and a thousand others. Most of their users have never heard the name. That’s the point. That’s what winning a layer looks like.
Agentic commerce is missing its layer.
The bottleneck is never the part you can see
When a new interface arrives, the flashy part gets the attention and the boring part is the actual constraint.
The agents are here. They can reason across a problem, weigh trade-offs, and hold a shopper’s context. The payment rails are here too — Stripe and Visa have both moved to let agents complete a purchase directly, which a year ago was the obvious blocker everyone pointed to. So the agent can think, and the agent can pay. What it still can’t reliably do is understand the products.
Product data is the seam nobody standardized. Every retailer publishes a different feed, a different taxonomy, a different update cycle. The richest detail about a product often doesn’t live on the product page at all — it’s scattered across reviews, a Reddit thread, an Instagram comment, a PDF spec sheet, an image with no alt text. A human can stitch that together by feel. An agent needs it resolved into something it can actually reason over: clean attributes, explicit units, normalized variants, a stable identifier that means the same product everywhere it appears.
Somebody has to do that work for the whole web, keep it current, and make it trustworthy. That somebody is the layer.
Why this layer is winner-take-most
Not every market consolidates around one provider. This one has the shape that does.
Start with the demand. The shoppers are already here — more than half of Gen Z uses AI to shop in some form, and that number only moves one way. The agents serving them won’t come from one dominant company; they’ll spring from a long tail of platforms, each with its own taste and interface. But every one of those platforms has the same underlying need: reliable, real-time, machine-readable product truth across the entire internet. None of them wants to build and maintain its own scraping stack. None of them should have to.
That’s the Plaid setup exactly. A fragmented demand side, a fragmented supply side, and one painful integration problem sitting in the middle that everyone would rather rent than build.
Then it compounds. The value of the layer is trust and coverage, and both feed on themselves. An agent learns which source it can rely on, so it leans harder on that source, so that source sees more of what agents actually ask for, so it gets better, so more agents rely on it. Coverage attracts agents; agents reveal demand; demand justifies deeper coverage. Like search, like Plaid, the first mover to earn real trust at real scale doesn’t just lead the category. It becomes the category’s default, and defaults are very hard to dislodge.
What I learned watching this happen once already
I’m biased here, because the unglamorous middle is where I’ve spent my career and it’s the layer I’m building now. So take the conviction with that in mind — but I came to it honestly.
I was an early data scientist at a payments company, and after it was acquired I spent a long stretch on merchant integrations. That work was deeply unsexy. It was also the moat. The product everyone admired sat on top of a mountain of normalization and reconciliation that no competitor wanted to redo. The flashy layer got the headlines. The boring layer got the lock-in.
Agentic commerce is going to rhyme with that. The agent is the part that demos well. The data layer underneath it is the part that’s actually scarce, actually hard, and actually durable.
In the next stretch of this race, someone plants a flag and becomes the product-data infrastructure the agents trust by default. Build the agent if you want the spotlight. Build the plumbing if you want the position.