- If anyone manages to build some truly fantastic software for massive numbers of end users, maybe those users will start expecting a higher bar for other software. Consider for example search engines pre/post Google. Naive search engines just don't exist anymore, because they were several orders of magnitude less efficient at getting useful responses. So that was a sea change. (Unfortunately monetization, SEO, sabotage and things like ignoring search keywords have resulted in search engines regressing in the last several years.)
- Long term design. Commercial software seems to be redesigned massively every few years, which is terrible UX. On the other hand, the vast majority of F/LOSS is designed once and either never changes again or just gets added to until it drowns in clutter. Even though the latter is caused at least in part by a lower budget, and often ends up being pretty bad UX-wise, I actually prefer it. One good balance is Amazon's web site. From a quick glance it looks like everything is where it was a decade ago, and overall it just looks familiar.
- UI convergence. Web browsers have converged so strongly that being able to do things like navigating and filling in a form in a completely unfamiliar browser is very likely to succeed. And browsers are so enormously popular that they have set expectations for how UIs work in other software.
- Quality assurance in all its myriad forms. A lot of energy is being spent touting the One True Silver Bullet, when in reality every approach has diminishing returns and shortfalls. If you use only manual tests every version is going to have more bugs than the last, because you don't have $MM to spend testing every single detail for every release. If you only use integration tests the suite is eventually going to run for longer than your release cycle and will be skipped and trimmed in ways which mean you'll miss bugs. If you only use unit tests you can only release fizzbuzz-size chunks reliably. And so on. Instead use every technique you know of until you hit similar diminishing returns for each of them.