Your next visitor won't be human

Translated from the original Spanish article.

It's a gradual, silent change. The internet is slowly turning into a space for AI agents, perhaps one day as much as, or more than, for people.

Here's an example: there are SaaS products that are “agentic,” or programmatic. This means they're used through a CLI or an MCP for AI, instead of the classic dashboard for humans. Recently, a founder explained that, despite having a base of tens of thousands of active customers, their dashboard (the panel designed for people) is barely used. So they've stopped prioritizing it.

In other words, the users of that SaaS are, essentially, other software or AI agents.

And it makes sense.

When we visit a website or use a tool's interface, we don't do it to kill time, but to look for something specific. A result. And that result almost always serves a purpose. Nobody goes into Analytics to wander around that labyrinth, but to find a specific figure that will help them decide.

If we extend this to the rest of the web, as I see it, interfaces are friction. Users have to read, navigate, search within them for what they want, and decide. UX/UI design often softens that friction, and even, in rare cases, makes it almost pleasant. But it can't eliminate it entirely, because ultimately it's an unavoidable path to getting what we're looking for. I particularly remember the book—and the idea—“The Best Interface Is No Interface.”

And this is where agents come in. An agent can go into the web service (Analytics, in this case) and remove the friction for the human, serving them the result directly.

For now, the vast majority of AI users don't work with agents; they talk to the model through the product's own chat. But that interface already accesses the internet and is connected to a range of tools.

Their online tasks are no longer just a matter of gathering information from a website: they go deeper with market research, look for suppliers, use tools, and even make purchases. Our own agents, for example, use paid APIs, such as Gemini's model for creating images or Semrush's for running studies; and, so far, the strict spending rules we've set for them have passed the tests. The agent that evaluates and decides isn't a hypothesis. It already exists, in its early version.

And I admit, gladly—this is important—that I need to go into the Semrush, Analytics, or Search Console dashboards, among others, less and less.

For now, the agency of agents (heh, heh. Sorry) depends on perceived risk. Take the case of OpenAI: in late 2025 it allowed purchases to be completed inside ChatGPT, with Walmart among its partners, without leaving the chat. It didn't work. Conversion was far lower than sending the user to the seller's website, and they withdrew the system a few months later. To me, it was a premature move more than a mistaken one. People still tend not to trust AI with purchases, just as they took a while to start buying online back in the day. But the trust threshold is rising, and the perceived risk of delegating more sensitive tasks to AI will keep shrinking.

The trend can be expected to continue, and we'll see the use of agents with more freedom and autonomy gradually become normal, until they carry real weight in purchasing decisions.

What this poses is a challenge of adaptation. And I believe the key lies in the when: deciding now, getting ahead, and not waiting for everything to be happening or for protocols to become standardized. Because agents are already roaming the internet.

And although most of them don't buy yet, they already do something that directly influences the buying cycle: they evaluate and compare. If a service isn't readable or accessible to them, it will be ruled out before the decision even reaches a human being.

That's why the first step is technical and necessary: removing the friction so that an agent can reach our website or service, navigate it easily, understand it, and use it. How that's done in detail is material for another post, and I'll bring it to you another day. Today it's enough for me to name the new paradigm.

Incidentally, around this new paradigm something is emerging that makes me very curious, something we don't yet know what to expect from and that no one can solve yet. Assuming the agent already finds us and reads us, what can we do to persuade it?

Decades of marketing have taught us to influence what weighs on people: 19.99 instead of 20, scarcity, social proof, loss aversion. But an artificial neural network may have other biases and other rules. It's quite possible that nothing from a classic like Robert Cialdini's “Influence” transfers to an AI.

So, for now, this is literally marketing for aliens: influencing a purchase decision without fully knowing what drives it.

With one caveat, which for me is the compass: however alien the new customer may be, some things don't change. The real quality of the service, the brand's reputation, and the quality of the product will go on carrying weight. No optimization can replace having something truly good.

I don't have the map of the future, and no one does. But I do have one conviction: this is quiet enough and technical enough that most people will ignore it until it's too late. And my hunch is that, for now, whoever starts making themselves readable to that alien visitor today will be the one AI recommends tomorrow.