I think today's LLMs and their derivatives (agents,.. ) are an impressive technological/research achievement with amazing real-world utility. Innovation at its best. I don't see the enshittification of commercial products based upon LLMs as taking away from that. Like I said, I see the potential of this technology in the local/open weights model space. Yes, those are currently noticeably behind the commercial offerings. But that's not a fundamental problem. It's not a race. If we keep improving open products they can one day match if not exceed the commercial options. A bit like open source desktop environments/operating systems - it took a while, but now the OSS options can arguably match if not beat the commercial ones.

For forms of income. E.g. with paid apps/gadgets/tools for everyday use (outside of dev tools) with chatgpt embedded.