The Next Attention Economy Has No Eyeballs · Ryan Merlin

writing about now connect<br>On this page Part I: The New Scarcity<br>What am I actually calling attention here?<br>Haven’t people already seen this?<br>Doesn’t a bigger context window just make this go away?<br>Part II: From Consideration to Action<br>Where does the money actually live?<br>What counts as the machine’s vote?<br>Where can you watch this happening right now?<br>Part III: Attention Becomes Capital<br>What does attention do to the thing it lands on?<br>What happens after a capability gets picked?<br>Part IV: Consensus, Concentration, and Control<br>If a hundred agents agree, how many witnesses is that?<br>Who owns the list?<br>Part V: Test the Thesis<br>What would make me back off the argument?<br>So what do you actually do about it?<br>The point

Over the last thirty years, the commercial internet has industrialized the fight for human attention. Sometimes companies buy it with ad budgets and prime pixels. Sometimes they earn it through search rankings, reputation, word of mouth, and the perfect headline. Both routes aim at the same target: a human nervous system that feels social proof, fears missing out, trusts a familiar name, and follows a crowd.

The next fight is already under way, and it’s stranger, because the buyer you now have to win over has no eyes. The old signals can still matter, but the nervous system they were built to move is gone. When an AI agent books the trip, picks the vendor, or assembles the software stack, it narrows the field to a few trusted options and drops the rest before any human sees a screen.

I build the systems where that choice gets made. After watching it happen a few thousand times, I can tell you it looks a lot like the old attention economy, moved to a room the customer never enters.

That old economy turned notice into traffic, traffic into data, and data into money. AI agents are opening a second attention economy upstream of the first. It begins before a person sees any options at all.

The first exchanges in this new economy don’t look like ad exchanges. They look like plugin directories.

Every enterprise buyer already knows this shape from procurement. Purchasing keeps an approved-vendor list. You can be the cheapest, most reliable supplier in your category, and if you’re not on that list, the buyer can’t cut you a purchase order. The list decides who’s allowed to be considered, long before anyone compares prices.

An AI agent keeps a list like that too, for tools and data sources and counterparties. This one updates itself, and nobody publishes it.

So here’s the bet this piece is making. The scarce thing in an agent economy isn’t notice. It’s admission to the set an agent is willing and authorized to consider. Admission is becoming rankable. The next question is whether it becomes ownable, then sellable. Plugin directories are the first visible exchanges, even if the real value ultimately sits deeper in the stack.

The argument in brief

Scarcity: AI agents create a new allocation layer between supply and human demand. The scarce position is no longer visibility alone, but admission into the set an agent is willing and authorized to consider.

Conversion: The economically important transitions are not listing or mention, but eligibility, retrieval, verification, recommendation, invocation, execution, and measured outcome.

Compounding: Repeated machine selection leaves residue. Successful tools become easier to retrieve, trust, authorize, and select again, creating a form of compounding attention capital.

Correlation: Apparent agreement across many agents may reflect shared models, registries, indexes, defaults, or infrastructure rather than genuinely independent judgment.

Control: Market power will increasingly belong to whoever controls candidate sets, permissions, defaults, and the evidence agents use to justify action.

Part I: The New Scarcity

What am I actually calling attention here?

Let me kill the ambiguity before the word runs away from us, because it points at four different things and only one of them is the subject.

I don’t mean the attention mechanism inside a transformer, the math that lets a model weigh some tokens more than others. I don’t mean a machine having an inner experience. I don’t mean the physicist’s observer. I mean attention in the plain economic sense the field has used since Herbert Simon, the sense in which a scarce resource gets allocated.

Call it allocative machine attention: the bounded consideration an AI system spreads across competing sources, tools, and counterparties when that choice changes what gets represented or acted on.

The definition earns its keep through three words. Bounded means the system can’t evaluate everything at equal depth, because it runs out of tokens, retrieval slots, latency budget, money, or human patience first. Competing means one option gets an opening another doesn’t. Consequential means the choice changes what gets represented or acted on: something gets surfaced,...