How I Finally Got My Local Environment Setup Right

I spend a lot of time thinking about the gap between what websites can do and what users can actually access. Most of my work involves building interfaces for humans—people who click buttons, fill forms, navigate menus. But over the last year, I've noticed that gap widening in a different direction. More and more users are delegating tasks to AI agents. They ask Claude to book a flight, tell ChatGPT to sign them up for a service, ask Perplexity to fill out a form. And most websites just... don't work with that. The agent can't see the buttons. It can't discover what actions are possible. It's like building a restaurant and then being surprised when people who can't see your door don't come in.

The frustrating part is that fixing this hasn't been straightforward. You could build custom API integrations, but that's expensive and fragile—you're maintaining two separate interfaces to the same logic. I tried various approaches before settling on a simpler pattern: drop in one script tag that automatically exposes what's already on the page through the WebMCP protocol, which agents understand natively. No rewrites. No API work. Just let agents discover your existing forms and buttons and widgets the same way humans do.

What I've learned is that this shift isn't coming gradually. It's already here. Every week I work on a site that needs to be agent-ready, and every week I see how quickly this becomes table stakes rather than a nice-to-have. The teams moving fast on this now—they're going to have an advantage.