The same is true in software. The first people in a company or on a project have by far the biggest impact on the structure and future of the software. People who join much later focus on smaller parts, and they might even be geniuses and coding superstars, but they will (naturally) work harder for narrower reasons on a smaller part of the project, relative to the whole.
Papers in established fields naturally have narrower and more specific, i.e., smaller impact over time because most papers are fine-tuning things, asking smaller questions, and not building brand new theories or frameworks from the ground up. It’s expected that Newtonian physics is not going to be re-invented every year, right? Newton did it, and now the questions left are how gravity works at scales and speeds we can’t observe everyday on earth. Nobody will ever supplant Newton, because he was first.
I’ve watched this happen in my own field, computer graphics. The early papers that have lasting impact are the ones that were inventing the field and laying the frameworks for how to think about it. The rendering equation, the shadow map and Phong shading are ideas that wouldn’t get published today, however they were pioneering at the time. Now the questions we have are about things like what is the true microfiber surface shape of human hair strands, so that we can increase realism by 1% compared to the previous hair models.
If you compare them side by side in the context of today, increasing realism of hair shading is a more difficult question to answer than the earlier question of how to interpolate a shading normal across a triangle.
So, yes scientific progress is waning in the sense that we’re inventing fewer fields and fewer new theories. There are fewer papers that are expected to be or even trying to be foundational, because the foundations already exist. And it’s not waning in the sense that scientific output has never been higher, and today’s papers are answering harder (and more specific) questions.