E.g. I had a very good experience in reversing a local bank API with LLMs to download my bank statements in a few seconds by local python scripts instead of several minutes of error-prone clicking in the bank's shitty old interface. The thing that I'd have done in one day, the LLM coded in several minutes by taking recorded request-responses. Yes, the code is a bit gibberish, but why do I care for my local single-user usage?
I can imagine a dozen similar stupid but routine API parsing challenges for LLMs that everyone could use.
If it's not enabled during usual browsing and doesn't snoop in everyday data, but only in a dedicated sandboxed window, I say it's a good design from Mozilla's side.