The Future of Commerce Is Agentic

For thirty years, e-commerce has been designed around one assumption: a human is looking at a screen. Product pages, carousels, “customers also bought,” checkout flows tuned to shave off one click. All of it is built for human eyes and human impulses.

That assumption is starting to break. More and more, the “customer” visiting a store will be an AI agent acting on someone’s behalf: comparing options, negotiating constraints, and completing the purchase. This post is my attempt to map what changes when that happens.

From search-and-click to intent-and-delegate

Today’s shopping loop looks like this:

1. I have a need (“I need running shoes for a half marathon”).
2. I translate it into keywords.
3. I browse, filter, read reviews, and compare.
4. I check out.

An agentic loop collapses steps 2 through 4. I state the intent and constraints (“neutral cushioning, under $150, arrives before the 20th, not the brand that gave me blisters”), and an agent does the rest. Sometimes it asks me to confirm, and sometimes it doesn’t.

The interface moves from pages to protocols.

The new commerce stack

When the buyer is software, every layer of the stack has to be rethought:

1. Discovery becomes a data problem

An agent doesn’t care about your hero image. It cares whether your catalog says, in a structured and trustworthy way, that the shoe has a 10mm drop and ships in two days. Merchants with clean, rich, queryable product data will win. This is where I think knowledge graphs quietly become a competitive advantage. An agent reasoning over “compatible with,” “replacement for,” and “made by a subsidiary of” needs relationships, not just rows.

2. Trust becomes cryptographic

If an agent can spend my money, three questions matter:

• Who is this agent, and who does it act for? (agent identity)
• What exactly is it allowed to do? (scoped mandates, like “up to $150, shoes only, this week”)
• Can the merchant prove the purchase was authorized? (verifiable intent)

The industry has already started building here. Payment networks and AI labs have announced protocols and programs for agent-initiated payments over the past couple of years. The common thread is the same: delegation with limits, and an audit trail.

3. Checkout disappears, and so does impulse

Much of conversion optimization is, frankly, the art of nudging humans. Agents aren’t nudgeable in the same way. If my agent is optimizing for my stated constraints, scarcity timers and upsell modals are just noise. Merchants will compete on actual value: price, availability, policy, and reliability. I think that’s a healthy shift.

What’s hard

It’s not all smooth. A few open problems I find fascinating:

• Preference elicitation. Most people can’t fully articulate what they want. Good agents will need to ask the right clarifying questions without becoming annoying.
• Liability. If an agent buys the wrong thing, who is responsible: the user, the agent provider, or the merchant?
• Adversarial merchants. Prompt injection in product descriptions is the new SEO spam. Agents need to treat merchant content as untrusted data.
• Agent-to-agent negotiation. When the seller is also an agent, we get machine-speed markets with interesting game theory.

What I’d build first

If I were a merchant today, I’d invest in three boring-sounding things:

1. A canonical, structured product graph, rich enough that an agent can answer “will this work for me?” without scraping.
2. Machine-readable policies: returns, warranties, and shipping SLAs as data, not prose.
3. An agent-friendly checkout path that accepts delegated, scoped payment credentials.

None of these are glamorous. All of them compound.

Closing thought

The web gave every store a storefront. Agents will give every shopper a personal buyer. The merchants who thrive will be the ones whose data is legible to machines and whose promises can be verified.

I’d love to hear how you’re thinking about this. Find me via the contact section.