If the software is actually good, it can start from an enthusiastic core of students, PhD students and later niche conference organizers and niche journal editors and if they gossip about their experience, it can spread through word of mouth if it's sufficiently good.

Yeah I think this is correct, with the added caveat that it must be as good as the alternative PLUS the awkwardness of switching to have any hopes of breaking out of the local minimum.

Otherwise you become Dvorak, which despite being better than Qwerty and having been around for almost a hundred years, still hasn't seen widespread adoption, in this case because the awkwardness of switching is very significant. The effect is likely smaller on something like Typst.