We picture agents as faster humans

We anthropomorphise new technology. We give it human traits and human habits, and that's why we imagine it doing the old job in the old way. It's also why we find it so hard to picture what a future with agents looks like.

The mental picture is a person clicking through a website, just faster and never tired. So we end up asking human questions, like how the agent should deal with the cookie banner or the captcha on a sign-up form. Both assume the surface stays and only the visitor changes. I believe that's the transition state, and one that will be around for a while, because most of the world's services still don't have an API. It's still a workaround, and nobody would build it that way on purpose. Our interfaces were shaped around human limits. We need a layout to read and a button to press. Take the human away and most of those constraints go with it.

So I'd stop asking how agents should use our interfaces and start asking what we would build if we'd never assumed there was a human on the other side.