In sciences, the real problem is things like the biology PhD whose training ends up being "working in this lab on this piece of equipment with this species for these tasks", or so I have heard from the other articles on this topic. And then there's the humanities, where a PhD is all but useless for anything that isn't teaching. There's varying levels of truth for this with other disciplines as well.
I'm not sure how much this is a widespread problem, to be honest. In preparing for my qualifying exam next month, and going to other peoples qualifying exam practice talks, there is a huge focus on making sure that people are asking relevant, interesting questions, and thinking through the analysis and interpretation of (as yet) hypothetical data. Those are skills that are broadly applicable, not just in biology specifically or science as a whole. People from my own lab have gone on, not just to academic jobs, but also science publishing and various startups (more and less biology related).
I don't think this is that big a problem. Publications are what science is made of. If you do research and don't publish it so that other people know about it, you're just engaging in a colossal waste of time.
You open-sourced your source code? Sorry, source code is not a publication. You improved your source code so that it is less buggy and acutally usable by industry? Sorry not a publication. You want to travel to PyCon to actually talk about your work to a wider audience? Sorry no travel funding available for such "non-academic" conferences. You want to cite an open-source work that did innovative work but not in academic setting? Sorry that is not a good enough citation ("who is this _why anyway?"), please include more citations of "academic" papers.
Funding incentive is only there for publishing your results but less incentive is there for building of useful working systems.
Results based funding (and tenure granting) promotes these more conservative and obvious projects. Proposals for more risky, and therefore potentially more rewarding projects aren’t submitted as no one wants to gamble with their own career.
I do think that the current PhD system is broken. As pointed out very correctly in a comment above: "I believe the problem lies in funding. Currently, we base funding largely off of publication success."
However, all is not lost. I have made a conscious effort in my PhD to not end up super specializing. I have tried to take up a problem that requires me to learn about multiple different computer science areas (compilers, architecture, machine learning and graphics currently). This has made my PhD a lot harder as I have to spend considerable amounts of time reading while others are publishing papers and getting better scholarships, I do think this is better for me in the long term.
Overall, while I no longer recommend anyone else to do a PhD, on the flip side if you work hard enough you might still gain value out of a PhD. Whether that value is enough is debatable, and requires considerable effort, but it is also possible to not super-specialize by picking your research topic wisely.
I do agree with the article in general and do think that more cross disciplinary stuff needs to happen. The academic system is currently largely unsympathetic towards trying to cover multiple sub-areas even within computer science with people in other sub-areas often being unwelcoming. Funding agencies are also unhappy with the much larger "startup time" spent in cross-disciplinary projects during which you spend time reading and not publishing.