The point about projects overestimating how well they know their users really stood out to me. The interviews show why direct feedback matters, especially in numerical JavaScript where something can run correctly without making its numerical validity obvious.

I also found the AI discussion interesting because the problem isn’t simply whether AI-generated contributions are good or bad. The bigger issue is that AI makes submitting code much easier while the cost of reviewing and validating that code hasn’t decreased in the same way. That puts more pressure on maintainers and existing review processes.

The finding about trust was also important. Stars and download counts can show popularity, but they don’t necessarily tell you whether a numerical library is actually reliable. Making tests, reference implementations, and validation practices visible seems like a much stronger way to build that trust.