Even though my focus is usually far from stealth mechanics for making my game, I still watched it in one sit because it's well documented and interesting. It really is hard to create a good balance.

Pretty thorough breakdown, and I'd agree with most of it. Some paraphrased bullet points though:

Perhaps stealth isn't doing well right now because of all of those bad stealth sequences in non-stealth games that are otherwise great

Possible, but I doubt it. The late 90s and early 00s was full of those too, but that was also simultaneously the height of stealth games. I think it's just that fads come in waves. Underground racing games were hot when The Fast and the Furious started. World War II shooters were popular after Saving Private Ryan, and modern military shooters followed when a lot of the target demographic either fought in the war on terror or knew someone who did. Dungeons & Dragons and a lot of Led Zeppelin's catalogue only exist because The Lord of the Rings books made a huge impression on people. If something in the cultural zeitgeist comes along that would make a good stealth game, there will be more stealth games; at this point, the last James Bond movie was 5 years ago, and the next one won't be for a while still.

A good compromise we landed on rather than perfect wallhack vision was to mark targets

Perhaps, but I'm even kind of sick of that, too. I played a good portion of MGSV without tagging and found it perfectly playable, largely because, as he acknowledges in the video, you can scope it out from any angle from far away before you make your approach. This is something the original Deus Ex and Ghost Recon: Wildlands do well too, even though the latter also does the trivialized target marking. I don't mind some light wallhacks when you're up close to simulate the character's hearing, like in The Last of Us. If we found ourselves in that stealth situation in real life, we'd have pretty good three-dimensional awareness of enemy locations just by the way sound bounces off the walls, and I think it's okay to represent that in some way. I think the marking target mechanic is so prevalent because Ubisoft made a large number of games featured in this video and inspired more still; and Ubisoft for the last decade and change has been a company that made an efficient machine for churning out games that used a lot of common mechanics between multiple franchises.

Playing through 007 First Light right now, a lot of my most common grievances are in predictability, but not of the enemies. There are a lot of context-sensitive actions, and while this game has a sort of tutorial that we'll be talking about for years, it did a poor job of explaining those contexts to me. I had similar issues blackjacking guards in Thief, and there's that guy-on-the-couch section in Splinter Cell that he highlights in the save scumming section that you could describe much the same.

While watching, I was thinking that roguelike could be a solution here. Cool to see that it's actually working out.

If I remember right, Cataclysm: Dark Days Ahead is also stealth-y + a roguelike. Gonna have to try that again. I believe, I actually found it too tense last time around, but might've just been past-me. 🫠

I also have to say, I really hate that mark-enemies-manually feature. As the wisdom goes: "Given the opportunity, players will optimize the fun out of a game."
Source

I'm that player. I will spend far too much time trying to mark any pixel of any enemy that shows up on my screen. It's not fun, but I'll do it anyways, because the game is telling me this is the big mission, this is where the prep work is important.

It also never made sense to me lore-wise. How can you be able to track people once they're marked, but you can't detect them without marking? Whatever technology you use to track their movement after marking, it has to be able to detect that movement without line of vision. If anything moves in the area, just show me that. 🤨

Gotta say, the alien in Alien: Isolation is my favorite "guard" in stealth games. That encounter in the medical area really is something.

One issue I've found with stealth games is that the concept of a lone commando outsmarting and sneaking past 30+ guards is pretty unbelievable to begin with, unless the player controlling the commando is actually a tactical genius. So, games tend to have to play pretty unrealistically in order to make the result playable by average players (idiot guards, x-ray vision, generous interpretations of darkness, super movement abilities, etc)

Something I think could be further explored is a game that explores a mechanic of "Plausibility". Stealth long accepted that guards would react to footsteps. Breath of the Wild including a noise meter helped me think "Hey, why is that? This is a big compound with a lot of people that's been quiet for the past week - wouldn't they assume the footsteps are their friends?"

Another concept that I think was explored in a very janky way but could use a second look is the game Shadwen. It's not a great experience, but it has the interesting idea of combining stealth with Superhot and Braid - you get full control over time, including time only moving when you do, and can rewind instantly from a failure. The game even shortcuts "spotted" phases by giving guards instakill crossbows, so the remedy to being spotted is just to rewind. The fun, then, is in creatively engineering the one-shot stealth circumvention you had in mind, even if a mispositioning messed it up the first 6 times (because it only took 15 seconds to get a successful effort).