majilesh
← All writing
AI & Agents··3 min read

When the Audience Is an Agent, Not a Human

Advertising has always assumed a human on the other end — someone to be persuaded, someone to click. That assumption is breaking. What happens to commercial infrastructure when the audience is a machine?

By Majilesh

For decades, online advertising has been built on one foundational assumption: a human is looking at the screen.

Everything flows from that assumption. The impression model, the click-through rate, the cookie, the conversion pixel, the A/B test — all of it is designed around human attention, human persuasion, human intent. You show someone something interesting. They click. You track that.

Now that assumption is starting to break.

What changes when the visitor is an agent

AI agents are increasingly the first point of contact between a user and the web. An agent searches for flights, compares software tools, researches vendors, and surfaces a shortlist — all before the human has opened a browser tab. By the time a person sees a recommendation, the "browsing" has already happened. It happened inside a model.

This changes the commercial equation entirely.

The click was never the point. The click was a proxy signal for intent. When an agent acts, the intent signal is richer and earlier — it's embedded in the query, the context, the user's stated goals. But there's no click. There's no impression in the traditional sense. There's a machine reading a manifest, evaluating an offer, and making a decision.

The manifest as the new ad unit

If you want to be discovered by an agent, you need to be machine-readable first. Not human-friendly first. The agent needs to understand what you offer, in what context it's relevant, at what price, and what action it should take.

This looks less like a banner ad and more like a structured data contract. A placement manifest that says: here is what I offer, here is my trust signal, here is the action endpoint to call if this matches your user's intent.

The advertising primitive shifts from impression-to-click to query-to-action.

Why attribution breaks

Current attribution is built on the click as an anchor event. Last-click, first-click, multi-touch — all of them trace a path back to a click. Remove the click, and the whole model needs to be rethought.

In an agent-mediated world, attribution needs to work differently. The agent selects an offer. The agent executes an action on behalf of a user. The conversion happens — but there's no click trail, no cookie, no browser fingerprint. What you have instead is an action ID, a delegation chain, and a conversion postback.

The durable redirect ID becomes the attribution anchor. Not the cookie. Not the pixel. An action identifier that follows the transaction through the agent's execution layer.

What this means for builders

If you're building commercial infrastructure right now, you have a window.

The companies that will matter in agentic commerce aren't necessarily the ones building the best AI models. They're the ones building the rails: the manifest formats, the trust signals, the action attribution primitives, the payment protocols that work for machine-to-machine transactions.

This is an infrastructure moment, not an application moment. The application layer will be built on top of whoever gets the rails right.

The interesting question isn't "how do I advertise to agents." It's "what does it mean for commerce to happen between machines, and what infrastructure does that require."

That question doesn't have a settled answer yet. Which means the answer is still being built.

#agents#agentic commerce#advertising#MCP#infrastructure
Share:𝕏 TwitterLinkedIn
← Back to all writing