From UI to Intent: WebMCP and the Rise of the Machine Native Web
I read the piece on the "Second Interface" and the Chrome WebMCP announcement slowly. Not because it was dense. But because it felt like one of those quiet shifts people shrug at… and then five years later swear they always believed in.
Let me try to unpack why.
The first interface was built for humans
The web was built for us : Buttons. Forms. Pages. Navigation bars.
Click. Scroll. Type. Submit.
Everything assumes a person sitting in front of a screen. Even when we automated workflows, we were sneaking around the edges. The real work happened through APIs. The web page was just the surface layer. That arrangement held for a long time.
But something is changing.
Agents aren’t just scraping pages anymore. They’re starting to show up as actual participants. And WebMCP feels like the browser admitting that out loud.
That’s new.
What WebMCP actually signals
If we strip away the launch-post energy and the core idea is simple:
The browser is turning into a place where agents can run, not just humans.
Up to now, getting an AI agent to use the web meant choosing between three imperfect options:
- Call APIs directly
- Reverse-engineer the UI
- Spin up a browser automation stack and hope nothing breaks
Most of us have done the third one. It works right up until a button moves or a class name changes. That fragility isn’t sloppy engineering. It’s the result of trying to use a human interface as a machine interface.
WebMCP points toward something cleaner: a structured layer between agents and web apps. Not scraping. Not duct-taped automation. An interface that’s meant to be used programmatically.
A second interface, running alongside the first.
Why this feels bigger than it looks
The first wave of AI adoption was chat. Ask a question, get an answer. The next wave is agents that take action.
They book things. Move money. Pull data. Trigger workflows. They don’t just respond; they operate.
Here’s the part I don’t think enough teams have internalised: if your product only exposes a human UI, agents can’t really see you.
It reminds me of SaaS in the late 2000s. You could ship without an API. Some did. But if you wanted to plug into the broader ecosystem, you eventually had to open up. This feels similar. Different actors, same lesson.
The browser as runtime
For two decades, the browser has been the runtime for human interaction. Now it’s slowly becoming a runtime for agents.
That shift changes what you optimise for.
Instead of asking, "Is this button obvious?" you start asking, "Can a system understand what this action does?"
That’s not a cosmetic tweak. It’s a design shift.
Humans need something that looks good and feels intuitive. Agents need something that’s explicit and predictable. Those are different constraints.
From interface to contract
Take a simple example: "Book a flight."
For a person, that means a form on a page.
For an agent, it means something else entirely:
- A destination field
- A date field
- Passenger count
- A payment endpoint
- A confirmation response
That’s not UX in the traditional sense. It’s a contract. You’re defining what actions exist and how they can be invoked. Some have started calling it AX (Agentic Experience)
WebMCP reads to me like an attempt to standardise that contract inside the browser itself. If that works, the browser stops being just a rendering engine and becomes a coordination layer.
That’s a different role.
What builders should be thinking about
If you’re building agent-driven systems, this probably feels familiar.The hardest problems haven’t been about raw intelligence lately. Models are capable. The friction shows up elsewhere.
Orchestration. Reliability. Dealing with brittle surfaces that were never meant for machine use. Connecting agents to messy interfaces takes more effort than making them smart in the first place.
If the web starts exposing machine-readable actions directly, some of that glue code disappears. Or at least shrinks.
Which means product teams will need to think in layers:
- A human interface
- A machine interface
- Some shared definition of intent underneath both
Most teams don’t have that mental model yet. But they will if agents start driving real transactions. Because once agents bring revenue, priorities shift fast.
The question I can’t shake
If agents become primary actors on the web, what is a website?
Is it still a destination you visit? Or is it more like an action surface other systems tap into?
In a world where agents do the browsing, traffic metrics start to look different. Visibility to agents matters. Clean action endpoints matter. Being composable matters.
Search engine optimisation (SEO) might slowly morph into something like agent optimisation. I’m only half joking.
A quieter kind of shift
Most meaningful changes don’t announce themselves with fireworks. They look technical. Slightly boring. Easy to scroll past.
But when the interface layer changes, everything built on top of it eventually follows. The first version of the web connected people to information.
If this plays out, the next version connects intelligence to action. And that means we’re not just redesigning pages.
We’re redefining who, or what, gets to use them.
Member discussion