Ask HN: Why are we still building web UI's just for human consumption?

1 points by ud0 ↗ HN
It is becoming increasingly clear that AI agents will do more on the internet and operate many websites, so why are web UIs still built only for how humans use them? Why isn't it being extended so AI agents can talk to websites? Right now, the ridiculous amount of complexity it takes for AI agents to operate a website and perform simple operations is crazy; look at the Claude Chrome extension, it has to take multiple screenshots and is so slow.

I built a React-like plugin that allows React-based apps to talk to AI agents easily using events. It can fill forms, react to alerts and stuff. This was very easy to do (full disclosure, I did a lot of front-end engineering at FAANG for a long time), and my AI agent was able to do stuff in seconds, way faster than most AI agents that operate the browser. This was easy to do, so why isn't everyone doing it?

6 comments

[ 1.8 ms ] story [ 20.7 ms ] thread
Why would I want my Ai using a web page instead of an API or CLI? Seems wasteful to spend tokens on image processing when a more deterministic, text-based alternative is/should be available
there are cases where an API is not feasible
than that is the exceptional case, not the default

I struggle to think of a case where it is technically impossible to have an API because the website is presumably making API calls under the hood

Right now, as we speak a lot of agents are doing clunky things to operate websites from Codex to Claude, all those website owners might not have the resources to setup an MCP or API just for agents. If this was an exceptional case how come Claude and Codex are still operating websites using clunk embedded web browsers and headless chrome complications.

It would take a lot to setup an API for certain resources when you consider stuff from security to compliance, but if you could just slap on a React plugin on already existing UIs so now agents can use the same site as humans already do.