5-minute read

Agentic commerce is a mirage. The future is Autonomous Commerce.

But why not agentic commerce? 2025 is the Year of Agents! Reasoning models are here. AI won’t just answer questions––it will do things for us.

But what things?

An agent could search for products we might like, choose them, and even place an order on our behalf.

But what if we enjoy the act of finding things? [See Chris Paik: driverless minivans v convertibles.] Or what if the agent buys something we don’t like? Can the agent make it right? Does it have the authority?

The agent won’t have authority unless it takes financial responsibility for the transaction. To do this, the agent would have to be making margin somewhere else, either by subscription or markup on an underlying transaction. The subscription is unlikely enough to cover mistakes particularly for larger transactions, and it’s unclear if consumers are willing to pay extra fees for a convenience of purchasing something through an agent instead of directly.

So if agentic commerce were not for the consumer, does agentic commerce help businesses? An agent could drive traffic to the business. But if an agent is just suggesting places to buy things and not actually taking action on your behalf, the agent is more likely to be a search or ads platform. Internet AI search is trying to act on your behalf just as much as a chatbot might be.

If it were to act on your behalf, the agent would probably need to disintermediate the brand – to replace the brand’s user experience and checkout with its own.

Most examples of agentic commerce are like this – they disintermediate the brand. Stripe’s Emily Sands explained to FirstMark’s Matt Turck last week:

We’ve all seen these cool demos of agents buying stuff for people: [At Stripe we have] this barista agent that is out there today, and you tell it what kind of coffee you like and it scours the internet for the best beans and it buys them for you. And what’s interesting about the barista agent is that it’s not a traditional coffee shop, it doesn’t own any of the inventory, it’s literally just doing the discovery matching + the payments – that is the entirety of the app.

But how does this app make money? And why do people use it over another aggregator or retailer with opinionated merchandising?

All this becomes clear by being pedantic about what a business actually is.

DEFINITION. A business is a wrapper over captured supply and its trusted delivery.

Emily goes on:

In practice agents have been buying for us for years: they were just human agents. When I order my salad from DoorDash, Doordash charges my credit card and issues a single-use virtual card to the driver who goes and buys the salad on my behalf.

This is where Emily trips up. While DoorDash drivers look like agents, they act as a part of an apparatus that contracts restaurant supply and underwrites every refund—so it’s a business, not an agent.

AI should at least be about saving consumers time and money. Today’s AI agents do neither. Operator is slow, expensive and unreliable.

So if not agentic commerce, what is autonomous commerce? Agents skip the hard parts; Autonomous Commerce embraces them:

DEFINITION. Autonomous Commerce is a system that owns, owes, and operates—owning the supply, owing you a fix if it breaks, and independently operates, giving users a limited interface to change its behavior.

Autonomous commerce embraces a view of agents as independent entities with their own goals and constrained interfaces to execute intuitive work.

  • Take me to Barry’s (I don’t care the route)
  • Pick up my groceries (I don’t care what car)
  • Deliver my burrito (I don’t care what robot)

Users of autonomous commerce don’t want to fuss with an agent’s intermediate steps. They probably don’t even care or want to know. Hiding away details is the point. The labor is conceptually self contained and grants users few knobs or dials to change the service mid-way.

This contrasts with agentic commerce, where an agent’s behavior is entirely contingent on upstream human values which are difficult to observe.

The simplicity and clear value proposition of autonomous commerce is why we’re seeing it scale relative agentic commerce, which has confused value propositions and complex tactics.

This opportunity is driving a new generation of startups shipping autonomous commerce to save consumers time and money:

While these Autonomous Commerce companies deeply integrate AI to provide enduring value, the most common agentic commerce use cases are feeling tired e.g., @near

i wish i had an agent to help me book a flight or order food, two activities i often need help with

or @signull

the tech is solid, & execution’s sharp but if i were openai, i’d ditch the flights & wedding demos. tired, overplayed, & utterly sterile. there are hundreds of easy to understand, richer, more human use cases that actually fit better.

 

the ones that are always used as examples don’t spark curiosity, they actually drain it. bordering on parody at this point.

While agentic commerce agents fumble through workflows where it’s not clear who’s driving who (see DeepMind’s @sebkrier exploration), autonomous commerce operates independently and is purpose built to hide away implementation details that actually don’t matter to the end-user. Their value propositions – saving consumers time and money – are clearly communicated and provide value relative to more expensive and fragmented human-led offerings.

Their combination of

  • Intuitive value
  • Owned supply
  • End-to-end liability
  • Well-scoped interface

makes me excited about the future of Autonomous Commerce.

Skip agentic commerce. Let’s make it autonomous.

Thanks to Ben Seidl and Mo Golshan for comments!