I left academia after my PhD. Here are a couple reasons why:

• The problems studied in TCS today are primarily of focus simply from previously being of focus, with no ground in eventual practical impact. In the field of data structures, for example, there have been a host of papers on the dynamic optimality of binary search trees, that is, finding the best BST in terms of visiting the fewest nodes for a sequence of element accesses ( Demaine et al., SICOMP '07, Iacono & Langerman, SODA '16, Kozma & Saranurak, SICOMP '19, to name a few), without asking why we are restricting to a binary search tree in the first place. Our work considered instead a general data structure for partial orders, a new angle which resulted in top tier publications—but garnering zero citations, as it is not what the community is focused on. At the end of the day, 42 paper folds get you to the moon, and practical efficiency must get beyond asymptotic analysis—O(log n) should be treated more like a factor of around 20.
• The academic community is ruthless. It took a year to publish our lattice paper (SWAT '20), since the paper corrected a mistake in an earlier paper (Talamo and Vocca, SICOMP '99). The authors of the previous paper were consistently recruited as subreviewers and denied the issue in their work, which was too technical for other reviewers (and perhaps the authors themselves) to discern. In another example, very little credit was given to our min-cut paper (SWAT '20) in Gawrychowski, Mozes, Weimann, ICALP '20 and Mukhopadhyay & Nanongkai, STOC '20, the former of which at times copied even the format of our paper. While communication with both sets of authors helped resolve the issues, to which I am thankful, the experience left for me a bitter taste of the incentives in academia.

For those wishing to pursue a PhD, I would first consider that there is typically much more to do in applications than in core science, but if one loves learning and is mathematically gifted, I would recommend pursuing it in artificial intelligence, where the applications are of huge practical importance and it is still (currently) possible to make meaningful intellectual contributions. This could easily change in a few years if we achieve AGI (as I suspect we will).