The Agentic Web Needs a Protocol

#webdev #ai #intentql #agents

"We built HTML for humans. But we never built a markup for machines."

Agentic browsers are coming.
Opera just announced Neon, their first browser that takes actions on your behalf. Others will follow. Everyone’s racing to build smarter interfaces that understand intent and can act autonomously.

But here’s the problem:

The web isn’t ready.
Even the smartest AI agents today are stuck scraping HTML, reverse-engineering buttons, guessing what elements do — and breaking every time a class name changes.

We’ve made browsers agentic.
But we’ve left the web opaque to agents.

Enter: IntentQL
IntentQL is a declarative, SQL-inspired protocol for describing how agents can interact with your site.

It flips the model:

Sites publish an agent.json file — like an API manifest for AI.

Agents send SQL-like queries to express what they want.

Responses come back as structured data — not HTML hacks.

sql
Copy
Edit
SELECT name, price, rating
FROM amazon.products
WHERE category = 'laptops'
AND price < 1000
ORDER BY rating DESC;
This isn’t scraping.
This is structured, intent-driven interaction — designed for agents, not eyeballs.

Why now?
Three forces are colliding:

Multimodal AI is better than ever at understanding human intent.

Agentic browsers are shipping (like Opera Neon).

The economic value of automating web tasks is massive — but blocked by fragile scraping.

We need a foundation.
A shared layer of intent between AI agents and websites.

This isn’t a product. It’s a protocol.
IntentQL is early. It’s not fully spec’d yet.
But it’s being built out in the open — designed to serve any agent, any site, any stack.

If you're a developer, a standard-maker, or just someone tired of fragile scraping…
You’re invited.

🔗 GitHub

🌐 intentql.dev

💬 Follow the journey — more soon