Writing

Jev makes structured decisions instead of generating text. I built a small Chrome extension to see how far that model can go when calling WebMCP tools.

How much of a workflow do people want to do themselves, co-pilot with an agent, or fully delegate? A framework for thinking beyond the demos.

If users bring their preferred agent to every website, businesses need a new way to represent their side of the interaction. Maybe the answer is a website-side agent harness.

Accessibility helps people use the web directly and gives browser agents a better foundation for assisting them - without taking control away.

A website starts a demo trip booking journey, a Chrome side panel finishes it, and nothing passes between them. What the 2026-07-28 stateless core makes possible, and what 'stateless' did and didn't mean before it.

Dynamic WebMCP tools keep the runtime tidy but hide the longer journey. Skills can hand agents a map - and Agentic Resource Discovery might help them find it.

On the wire, delegating to a page's agent looks identical to calling a plain tool. The difference is who does the thinking - and, it turns out, whose wallet pays.

In-page, in-browser, off-browser - where an agent lives decides who holds control, how far a task can travel, and whether a human ever sees your website.

A working definition, why the agentic shift is real, and what it means for the sites we build.

Explorations, demos, thinking out loud