You are missing that you actually pay a lot for those abstractions. Once you get to Twitter's or Instagram's scale, it starts to pay to DIY the foundations of your technology, because the people to build it are cheaper than the cost to buy it. Building also has the benefit of aligning your concerns across abstraction layers, which pays huge dividends in efficiency.

I don't think it's obvious at all that the long term trend is in favor of few employees with "high impact through abstraction" when the highest-impact people I know at big tech companies are ones who build the underlying technologies behind those abstractions. They routinely bring in 8-figure cost savings per engineer per year compared to using OTS or open-source technologies.

Headcount should scale with actual complexity of the problem being solved, which is not AT ALL linked to a user's perception of "complexity." Twitter is far more complex than you are giving it credit for, and a big part of that complexity comes from its scale and the fact that it is cheaper for them to eschew "higher level abstractions" than to use them.