Recently, Anthropic and Google have announced that their agents can now interact and operate a browser on a user's behalf.
As impressive as these capabilities are, something about this feels off to me. Instead of building agents to interact with the internet as people currently do, shouldn't we be designing agents to interact with the internet in an agent-native way?
Websites are representations of information, and they are rendered with people in mind. That's why we have big bold copy above the fold, a clear CTA, good spacing between elements, visual elements to add interest, etc.
The thing is, the same information can be easily represented as a .txt file or as a JSON or as a RSS feed or as an API.
Instead of taking a human-centered design approach to designing websites, what would it look like if we took an agent-centered approach?
What would usability and desirability even look like in this context?
Will everything be shuffled into a renewed robots.txt file?
Will everything behave like a market?
At best, I think of these examples as stepping stones to the frontier of the new B2B - Bot to Bot - economy.
I think it's neat that agents can control the browser, but it feels like a strange ouija board experience. Do we really need to show the mouse sloooooowly moving across the screen or can we fast forward to the agent negotiating and completing a purchase of a Japanese mini truck on my behalf?
What's clear is that we're designing a way for agents to interact with our version of the internet. We'll see if they will design a way for us to interact with theirs.
