I watch MrPete222 on YouTube. He's certainly capable of whipping things up on the lathe on occasion without drawings, and he's no novice, but it's common for him to spend days or a week making a part that you could design in CAD in fifteen minutes. Same with Abom79, Stefan Gotteswinter, Joe Pieczynski, and Tom Lipton. Dan Gelbart is still using the spot welder he made in his youth 30 or 40 years ago.
Some of these guys "eat engineers for breakfast" but they still can't "wake up, ... be presented with a new problem no one else has addressed before, and build a machine to solve it, all before dinner." The timescale for solving novel problems just seems to be a lot longer. Even building a steam engine from a kit is a matter of weeks on a manual machine.
If you need a key copied, a lock pick ground from a street sweeper bristle, a transfer punch made from drill rod, or a weird left-hand screw with a strange thread pitch, they can probably do that before dinner, yeah. But that's because those are well-understood and fairly simple problems.
CNC machining opens up more possibilities, but the timescale is still longer. You don't want to find out your G-code is buggy by crashing a US$150k milling machine; crashing a mill is more like crashing a truck than crashing a computer. So machinists are generally methodical, patient, and careful.
What in particular are you referring to that could have been designed on CAD in 15 minutes here?
Normally, whenever I have seen an Abom79 video doing something, it was some big-ass part that they were doing manually because either:
A) they were refurbishing something so they had to kind of cope with any non-idealities that arose
B) the part was very simple but was just a really big-ass part that really didn't fit well on anything so that had to kind of make it up as they went along.
Of course, it could just be that I don't really watch Abom79 videos unless it's about some big-ass part, so I could very well be biased.
Yes, quite commonly he spends several days making a part where he gives you all the dimensions and tolerances in three minutes at the beginning of the video, and the reason it takes several days is that if at any point he screws up, the part will obliterate him, his lathe, both, or his client, so he doesn't rush anything. And, yes, he spends a lot of his time coping with non-idealities. These are among the reasons that whipping up a quick Perl script is so much faster than whipping up a quick hydraulic cylinder.
I'm getting a taste for that since I got into 3D printing. Just type in some math formulas in a text editor, press a few more buttons, and all of a sudden I get a pivot-fence micro adjustable router table for my Dremel tool. Or I get a doo-hickey that holds a nut and washer, and can screw under my full sized router table to hold the nut in place so I can screw in various pivot-posts or other 3D-printed attachments on the top side. Or make a custom connector for something on my bicycle. Essentially I can fabricate products that just aren't purchasable.
But I really wish I had machine shop skills, as there is so much more you can do with a lathe, stamper, roller, and other equipment with basic input materials.
You may enjoy reading the non-fiction book, "Hot Tech, Cold Steel", by Chuck Hutchins. It's his story of starting a tech company on the forefront of CNC machining.
One of the chapters talks about Chuck's experience in his first job after graduating from college. He accepted a relatively low compensation offer at a machine tool manufacturer, on the condition that he would spend his first year sequentially shadowing every other employee. There's an anecdote about him being handed a power tool by a machinist and told to go ahead and cut metal on a several hundred thousand dollar linear rail (in 1950s dollars), and the boss running and screaming across the shop to stop him.
You've never met that wizard with the machine shop, have you? He'd be insulted if you called him an engineer, but he eats engineers for breakfast.