The most common objection I hear to building a cleaning-and-structuring layer for AI agents is some version of: won't the models just get smart enough to read raw HTML themselves? It sounds reasonable. It's also wrong, and the reason has nothing to do with intelligence.
The models-will-just-handle-it fallacy
The assumption is that capability is the bottleneck. Make the model smart enough and it parses any messy page, ignores the nav, finds the price, moves on. Models are already that smart. You can hand a frontier model a 2 MB HTML blob and it will usually find what you asked for.
But can it and should it are different questions, and the second one is decided by economics, not capability.
Tokens are the unit of gravity
Every token an agent reads costs money and latency. A raw web page is mostly tokens you don't want — scripts, inline styles, tracking, navigation, boilerplate. Feeding that to a model means paying to process noise, on every request, forever.
Clean, structured input can be a tenth of the size of the raw page. Same answer, a fraction of the cost, a fraction of the latency. Run that at scale — thousands of fetches an hour inside an agent loop — and the difference stops being a rounding error and becomes the whole bill.
Intelligence doesn't repeal economics
Here's the part people miss: making the model smarter makes this worse, not better. A more capable model is more expensive per token. The smarter the frontier gets, the more it costs to waste its attention on a cookie banner. Cost gravity pulls in the opposite direction of the it-will-just-handle-it argument.
Capability is not the constraint. Token economics is — and economics always favors the cheaper format.
The compatibility layer is the cheap path
This is why a structuring layer isn't a temporary crutch that smarter models will retire. It's the economically correct place to do the work. Clean once, serve cheap, and let the expensive model spend its tokens on reasoning instead of parsing.
Bigger models don't fix the format problem. They make solving it more valuable.