The abstract did say the result is mixed. You have "long term" increase in human capital development...primarily because connections help mentor more junior developers, but output is reduced...for obvious reasons.
The "output is reduced" especially for certain crunches where time is of the essence IS an argument for WFH in those circumstances, and for me, when I need the most time alone.
The abstract says one thing, the title here suggests an entirely different thing. Besides that not-so-subtle editing, I also find the sample size used more than a little bit lower than one that you could draw such a sweeping conclusion from.