I'm sure Leetcode and similar exercises are great recreational activities, whose benefits leak into daily work in some shape or form. The problem I see is that the practice has been manifested as a one-dimensional interviewing practice:
"Solve this problem in 45 minutes. Top to bottom, single pass, no typos, right now, while I'm watching and interrupting your thought process. You are not permitted to feel anxious. You now have 43 minutes - Go!".
I understand there are companies which treat Leetcode as a "representative field test", but IMHO (based on my 8+ yrs experience working across FANGM, w/ ~200 interviews conducted) any company that uses Leetcode this way, is doing it wrong.
The "right" way to use Leetcode, is to treat it similarly to how IBM does their weird IQ test screening, except Leetcode is strictly better than an IQ test. It is NOT a simulation of work. It is also NOT an "intelligence" test (thank god for that).
If anything, it is closer to how professional sports teams evaluate young athletes via a training combine or a set of individual drills/tests. Doing vertical jumps or sprinting really fast are not "representative field tests" of how well you actually play a sport (the whole sport, not just a single part of the sport), but they are good-enough filters for an athletic "interview".
However, in the same way that IQ tests are bad for a multitude of reasons including implicit demographic biases, so too are Leetcode tests flawed (including a heavy element of age-ism, as there is a strong anti-correlation between "years since college" and affinity for Leetcode-style tests).
"Solve this problem in 45 minutes. Top to bottom, single pass, no typos, right now, while I'm watching and interrupting your thought process. You are not permitted to feel anxious. You now have 43 minutes - Go!".
This is not a representative field test.