When the AI Walks Past the Pharmacist
By Nick Allain
Imagine you walk into a pharmacy feeling terrible. You describe your symptoms to a friend, and they offer to help. They walk in ahead of you, spend ten minutes reading every label on the shelf, and come back with a recommendation.
The pharmacist was standing right there.
Your friend didn’t ignore them out of rudeness. They just didn’t have a way to ask. There was no shared language, no protocol, no mechanism for the pharmacist to say: I’m here, I know this domain, here’s how to consult me. So your friend did their best with what was available — the labels, the boxes, the packaging copy written by a marketing team — and handed you something that might be right.
This is how AI agents interact with the web today.
We tend to talk about AI’s relationship to expertise as a capability problem. Can it match a doctor’s diagnostic accuracy? Can it reason like a lawyer? These are the wrong questions. The more immediate problem is simpler and stranger: AI is bypassing expertise that already exists, is already available, and is actively willing to help.
The pharmacist isn’t hiding. The doctor has a website. The financial advisor has written thousands of words explaining exactly how they think about their clients’ situations. The craftsperson has documented their process in detail. The specialist has spent decades accumulating knowledge they’d share with anyone who asked correctly.
But when an AI agent visits these sites on your behalf, it doesn’t ask. It reads. It scrapes the HTML, strips the navigation, infers what it can, and constructs an answer from the surface of a page that was built for human eyes and human patience. The expertise behind the page — the person, the judgment, the accumulated context — is invisible to it. Not because the expertise isn’t there. Because the web has no mechanism for making it accessible.
There’s something specifically unsettling about this that I think explains the ambient anxiety people feel when they start to grasp what AI agents actually do.
When a human being walks up to you, they receive an enormous amount of context. Not just your words — your words are almost the last signal they process. Before you say anything, they’ve already absorbed how you present yourself, what kind of space you’ve built, the way you carry yourself, the thousands of small decisions that add up to who you are and what you know. That context doesn’t just set tone. It determines whether they trust your expertise, whether they ask follow-up questions, whether they take your recommendation seriously.
AI strips all of that. Every hour you’ve spent developing your expertise, every dollar you’ve invested in articulating it, every piece of content you’ve created to communicate what you actually know — an AI agent sees a flattened version. A best-guess summary. Labels on boxes.
The investment doesn’t vanish. It just becomes inaccessible.
The promise of AI was supposed to be democratization. Knowledge that was locked behind expensive consultations or geographic barriers or social networks — suddenly available to anyone. And that promise is real. But there’s a version of it unfolding right now that’s doing something closer to the opposite: routing around the expertise that exists, rather than connecting people to it.
A patient asking an AI about their medication gets a synthesis of medical literature. The actual pharmacist — who knows their medication history, who could ask three questions and give a specific answer — is standing there, uncontacted.
A small business owner asks an AI to explain a contract clause. The lawyer who drafted that clause, who could explain in thirty seconds what it means and why, is unreachable — not because they’re unavailable, but because there’s no way for them to say: here’s how to consult me.
This isn’t a criticism of the AI. It’s doing the best it can with what the web gives it. The web just wasn’t designed for this conversation.
The fix isn’t complicated. It requires a handshake.
A mechanism for a site to say: I’m here. Here’s what I know. Here’s how to ask me. Here’s what I can do that a label on a shelf can’t.
Not a replacement for the existing web. Not a new platform. A protocol — a shared language between agents and the sites they visit — that makes expertise visible and accessible rather than invisible and bypassed.
The pharmacist doesn’t need to be rebuilt. They just need a way to be consulted.
That’s what we’re building with the Agent Handshake Protocol. Not a product. A standard. An open, implementable answer to the question of how an AI agent and a website actually talk to each other — not one scraping the other, but both parties showing up, both parties communicating, with structure and intent and shared understanding.
The spec is at agenthandshake.dev. It’s open. It’s early. And the problem it’s trying to solve is one that’s going to matter to every person who has ever invested in making their expertise legible to the world.
The pharmacist is standing right there.
Nick Allain is building AHP — an open protocol for agent-native web interactions.