Skip to content

What Operator Shows Us About Tomorrow's Web

Published: at 12:00 AM

I’ve been playing with OpenAI’s Operator Agent for a few days now, and it’s got me thinking about how broken the web is for AI.

Here’s what fascinates me about Operator: it just doesn’t quit. You know those annoying web tasks that make you want to throw your laptop out the window? Filling out endless forms, navigating garbage UIs, hitting dead ends when trying to talk to sales? Operator just methodically powers through them. No frustration. No distractions. No giving up.

It’s like watching someone try to eat soup with a fork. Sure, they’ll eventually get there, but it’s painful to watch. That’s Operator trying to use today’s web. It works, kind of, but it’s clearly the wrong tool for the job.

Right now, everything on the web is either built for humans to click around or for machines to follow strict rules through APIs. That creates some pretty obvious problems:

  1. APIs are too rigid – you have to predict every possible thing someone might want to do
  2. A lot of cool tools never get built because making robust APIs is a massive pain

Don’t get me wrong – the human web isn’t going anywhere. Neither are traditional APIs. But watching Operator work makes it obvious that both were built for a different era. We’re going to need something new for the AI-native world we’re heading into.

What if we built web services specifically for AI agents? Instead of trying to plan for every scenario upfront, services could be flexible and adaptive. They could learn how to interact with different agents on the fly.

A few things would probably look pretty different:

Operator is giving us a peek at what’s coming. Think of it like adding a new layer to the web – one where AI agents can interact in ways that would never work for humans or traditional machines. The web is about to get a lot more interesting.