Interesting ideas. So you're learning more towards the recommendations (and eventually agent) use-case. I find having to provide ChatGPT / Claude with context before prompting to be a real inefficiency and them having access to your online history would streamline this more.
Oh yhea indeed, It's super anoying to tell the AI every damn time that I don't want it to write me python code for every example !
Which makes this an excellent question. Indeed perhaps the most important question to ask before you build anything.
Too many programmers subscribe to the school of "if you build it through will come'. (Hint: that's not how it works.)
Firstly you don't have to show anything to potential customers, (and even when you have it, it's better not to show it.) You are selling benefits, not features.
Secondly, how to find them is exactly the hard problem you need to figure out (it'll be different for different products.) For me personally, I the late 90s / early aughts, I traveled to user groups (all over the world.) Turns out that "showing up" spread the word very quickly.
(That worked for me because it's a niche area and nobody else bothered to travel, so I stood out.)
The key lesson here- if you can't define your market, and if you can't reach them, you can't sell to them.
Every business has hard problems. This may be yours. Solve this problem first, not last.
This quote really stod out tho: If you can't find people before you make it, Then you won't find them after you make it either.
literly every time i just wasted money in business was because i didn't work by this quote!