fleeting and very much WIP, possibly a brain fart
I use DuckDuckGo or sites like marginalia.nu as dumb text indexes of the web, then LLMs (with or without web access) to synthesise answers or solutions. Google and friends sit somewhere in the middle, with a vague distinction between both approaches and being mediocre at either.
When asking an AI a question, I have a stronger having of double checking the answer, because my bullshit detectors are set to high. When I need a response that's just good enough. When using an AI to answer a question, I'm more likely to follow-up by reading a paper or opening Wikipedia.