Idea – machine learning algorithm that browses the web?

Are there companies working on a machine learning algorithm that browses the web?

I was recently summarizing a lot of resources on a website that are not presented in a structured way - some videos, some audio, some PDF, some summaries, no clear schema, no RSS, definitely no "download all" button - that were only accessible behind a paywall and login. It occurred to me that a browser extension could be used to navigate this information and run it wholesale through a local GPT-type model. This browser extension could use behavioral data to "learn to browse" and detect user interaction (such as video elements watched, rewinds, semantic value of content presented). Surely some company is working in this space - who?