Hardcore manual memory people seem to have completely misaligned priorities to real-world concerns. Maintainability, safety, productivity, and iteration speed seem to be bottom-of-list to the egotistic holy grail be being able to say they write their own memory routines. If they're not in an embedded context, I'm really unsure why anyone paying them to write code should care what they have to say. They're not talking about increasing robustness and reducing costs, so what the hell are they even rambling about? I've had too many of these debates in-person face-to-face with people who can't match my ability to deliver.
The vast majority of us are not smarter than the compiler, and are not smarter than automatic MM routines. Those who are write compilers and GCs. They don't work on the same programs we all write, just memory managed. It's the worst-kept secret that the remaining contexts where manual management is used often have the worst-maintained spaghetti codebases, leading to disasters, whistleblown scandles, and hush-hush covered catastrophes waiting in the wings while people get the hell out of dodge before they blow. It's all duct tape and prayers. Having had to inspect and translate procedures from a deliberately obfuscated spaghetti C codebase, my position on this is going to be hard to budge. Experience is an unbeatable Bayesian prior.