Applying to PhD Programs in Computer Science (2014) [pdf]

ᴛⷮyрⷬoͦ iͥn ᴛⷮнⷩEͤ 1ˢᵗ ¶ 𝕒𝕗𝕥𝕖𝕣 𝕚𝕟𝕥𝕣𝕠𝕕𝕦𝕔𝕥𝕚𝕠𝕟.

Person A never does any serious training in a sport, and then goes on to spend 10 years playing "hit and giggle" tennis (say). They probably get a bit better.

Person B spends several years in their youth doing tennis lessons at a high level. They then never take another tennis lesson, ever, but they spend the same 10 years as person A playing recreational tennis, using the skills they have built in their youth as a framework.

I feel that a lot of the stuff I've done recently builds more on the work I did as a commercial researcher (2006-2017, particularly) than it does on my PhD work, but I also think that the 2006-2017 work greatly benefited from my PhD.

I applied to 6 schools - doing the "bracketing" thing (CMU+UCB as "reach" schools, Wisconsin-Madison & Washington as "mid" schools and Columbia and U Colorado Boulder as safety schools). I got into everywhere but UCB and Washington, but CMU was my first choice so I was super-happy.

The thing was - there were specific people at Boulder and Columbia that I thought I wanted to work with, but I checked a couple years later out of curiosity, and they'd moved on (around the time I would have been doing a thesis proposal). My first adviser at CMU also went to industry after a couple years and I had to switch advisors and fields, but had a number of excellent choices - as top-tier schools are packed with really good people.

So if your interests shift, or you are forced to have an advisor change, or you wind up working in an area with some adjacencies to your original area (say you start as a networks person but wind up doing a half-networks, half-operating system PhD) you will find good advisors and coadvisors aplenty at a good school.

At an "OK" school there are usually some really good people but they may have zero good people in whole subfields, and you might find that that good academic has left before you start your PhD proper...

Agreed, a PhD is absolutely not useless in this industry, even if you don't end up in the research community. Understanding where the research frontiers of various fields are and being able to quickly find / digest relevant technical papers feels like a superpower. The gap between an undergrad education and a research frontier is enormous, and only working on a PhD really gives you the time and incentive to cross it. Having done it once, it gets easier to do it again.

For me, it has turned a huge volume of "unknown unknowns" into "known unknowns" and equipped me with the tools to then convert those into "knowns". Without it I'd be a fine coder, sure. With it I can work on a different tier of projects, and direct my career much better.

The costs are very real, though. Giving up ~6 years of early career earnings in a high-paying industry is utterly insane; you will never, ever make it up short of your startup lottery ticket number coming up. It's a meat grinder for mental health. Dozens of things outside of your control can go wrong and torpedo your aspirations. It is the right choice only for a vanishingly small minority.