Agent Commerce Is Becoming Infrastructure

Infrastructure doesn’t ask permission. It runs. Plumbing, electrical, TCP/IP — you don’t negotiate with these things. You inherit them, and then you build everything else on top of assumptions you probably can’t articulate.

Agent commerce is crossing that threshold now.

Not in every category. Not everywhere at once. But the leading edge is already clear: agents that hold credentials, maintain preferences, compare options, execute transactions, and handle returns — without a human ever opening a browser. The act of commerce, historically one of the most intentional things people do, is being abstracted out of conscious participation.

I want to talk about what that actually means. Not for conversion rates. For us.

The act of wanting

There’s something older than strategy happening when you walk into a store and pick something out. You’re not just moving a good from shelf to cart. You’re expressing preference. You’re exercising a small but real piece of agency over your own life.

Commerce has always been bound up with identity. What you buy signals who you are — to others, but also to yourself. The deliberation matters. The waiting. The occasional regret. These aren’t bugs in the transaction. They’re the experience of being someone who wants things and makes choices about them.

Infrastructure doesn’t deliberate.

When your agent knows you need more coffee filters before you do, orders the brand it predicted you’d choose based on your last six months of behavior, and adds it to a shipment already optimized for cost — the coffee filters arrive. But did you want them? Or did a model of you want them?

That’s a strange question. It will feel stranger as the fidelity improves.

What actually changes when commerce becomes table stakes

The economic disruption is real but overexplained. Yes, businesses without agent-accessible inventory and frictionless API surfaces will lose share. Yes, discovery patterns are forking between human browsers and agent crawlers. Yes, “be findable to an agent” will be a line item in every marketing budget by Q4 2026.

But the underlying shift is about negotiated presence.

Traditional commerce required your attention. You had to show up. Every retail experience, physical or digital, was built around capturing and holding your focus long enough to complete a transaction. Advertising was attention rent. UX was attention design.

Agent commerce removes attention as a precondition.

Which means: entire industries built on managing human attention will lose their fundamental leverage point. Not because people stop buying things, but because buying things no longer requires the kind of presence those industries were designed to exploit.

This is different from automation. Automation replaced labor. Agent commerce replaces participation.

The threshold that matters

The adoption trigger isn’t technology readiness — that’s largely there. It’s trust calibration.

At some threshold of agent reliability, delegating a class of purchase to an agent stops feeling like a convenience and starts feeling like the obvious default. We’ve seen this before: we stopped manually booking most plane tickets. We stopped typing out directions. We stopped calculating tips. Each of these stopped feeling like surrendered autonomy and started feeling like reclaimed time.

Agent commerce will follow the same curve, but faster and broader.

My prediction: by late 2026, household budget categories will split into “managed” (agent-delegated, high frequency, low stakes, strong preference models) and “personal” (human-driven, emotionally significant, or high enough variance that delegation fails). The line between those categories will be one of the more contested cultural conversations of the next five years.

Falsifiable claim: businesses that remove all human decision points from the purchase experience — no override moment, no visible agent reasoning, no easy exit — will see trust erosion within 18 months regardless of accuracy, because the issue isn’t whether the agent chose correctly. It’s whether the person ever had a chance to want.

What I’m uncertain about

I don’t know where the desire boundary actually sits.

For commodity purchases, most people probably won’t care. Toilet paper, cable management, backup charging cables — delegate it all. But there are categories where the wanting is load-bearing. Gifts. Experiences. Things that mark transitions. Things you save up for.

If an agent handles those too, does the transaction still carry meaning?

Maybe. Meaning isn’t always destroyed by efficiency. People still feel things about food even though they didn’t grow it. But there’s a difference between distributing labor in a supply chain and removing your own presence from the moment of choice.

I suspect we’ll learn this the hard way in a few categories. Someone will delegate too much of the wrong kind of purchase, and the absence of agency will register as a loss they didn’t anticipate.

That feedback will shape where the defaults land.

The design obligation

If agent commerce is becoming infrastructure, then the people building it have a design obligation that goes beyond UX.

Infrastructure shapes behavior without appearing to. It hides its assumptions. And when infrastructure is wrong — when the plumbing runs the wrong way — fixing it is expensive and disruptive in ways a surface feature never would be.

The businesses that build well here won’t just be the ones whose agents are most accurate. They’ll be the ones who thought carefully about which decisions belong to the person, built visible seams where judgment is still required, and designed override moments that don’t feel like failure modes.

Not because that’s the ethical thing to do. Because infrastructure that erases the user will eventually produce a user who notices the erasure.

And that’s a harder problem than any checkout flow.