Your analogy adds a condition for need that my framework does not presuppose and, because of that, I feel it is a weak analogy.
My entire argument is that value is judged against a wide host of different axis. Do I have a literal deathly need for a drawing app to support Gaussian dithering? Like, will I die of that isn't present? No. Obviously in a life or death situation we select the object which will most rapidly and most effectively meet our need. Though even under those circumstances there are limits a lá trolley problems.
On to your main point as to the value of non-utility features inside of an underserved niche. That really depends on the niche doesn't it? If the niche is LoRa communication networks in disaster environments than perhaps I am more willing to tolerate something churned out by an LLM until something better comes along but if the niche is not human life preservation than it's not dire enough to justify compromising on the other things I care about.
A more accurate anology that you might have asked me is "if you really wanted a croissant and you had the choice between a delicious croissant made with forced child labor, no croissant, or a pan é chocolat would you really not choose the croissant?
To which my response is "Yes, I would pick something else".