29 minutes ago, Fulcrum.013 said:

Assume that  99% of  data needed to phisic engine e.t.c modelling software able to compute from geometry, other 1% is manually inputed into modelling software. Also it helps to build geometry at least 10x faster than usual modelling tool. So you have about 99% of its work done automatically. 

That won't help you if the physics data isn't related to the object's physical geometry. That is quite often the case. The character physics in a first-person shooter rarely has any kind of (mathematical) relation to the character models. Usually the parameters are hand-tweaked by designers until they feel fun. To the best of my knowledge, tweaking for fun cannot be automated, because "fun" refers to the player's emotional reaction to the system as a whole. The purpose of a video game is to elicit that emotional reaction. How do you automate a process that necessarily involves human emotion to evaluate it?

There's no physical reason why the TF2 Scout would run so much faster than the TF2 Sniper, for instance - Sniper is taller and has longer legs and generally looks more athletic, yet Scout can not only run and jump much higher than him, Scout can also *double-jump!*

10 minutes ago, Lactose said:

As long as there is a manual process involved in creating games, AAA studios will always be able to stand out by throwing more people/money at the manual process.

21 minutes ago, ChaosEngine said:

The more amazing your engine is, the more work has to go into the asset production to show it off. Sure, tooling in that area has improved, but it's still a very labour intensive process and it turns out that level designers, modellers, etc like to be paid. 

8 minutes ago, Fulcrum.013 said:

Just take a short look to industrial modelling tools that intended to make a data for physics simulations from geometry.

Fulcrum is right, but those things seem to be widely unknown to gamdev i guess (because it's not all mature yet - that's one area where we have to progress.)

Personally i came across this stuff unintended. I work on real time GI which creates some form of lightmaps, and i wanted seamless UV maps to apply this to game geometry nicely and effective. This is the same problem as quadrangulation, which is a heavy industrial design research topic for a decade with constant progress, but no good enough solutions do exist as of now, at least not good enough for me.

However, while working on this i realized this has many other advantages as well. Automated LOD generation, seamless lod blending at runtime, proper displacement mapping, all this comes as a bonus. Finally you have faster rendering with more details, simply better geometry, easier texture space lighting, and: MUCH less work for artists (they only work on very high res and care a shit about polycount, texture budgets, UV maps or LODs - except for things like characters likely)

That's still lots of work, but it is possible. I would know nothing about this by looking at how current AAA games are made, which is the point i try to make, or at least one of them. (not the first time i mention this, but it fits here. I took me much more time than a short look to get the vision.)

16 minutes ago, Fulcrum.013 said:

Of cource logic of controlled characters usualy tweaked. But it can be done in minutes and tool to tweak it can bee implemented in hours. But it millions of other tiny details that have to look realistic even into complete fictional FPS. And it require a accurate data for simulation. 

The actual tweaking part can be done in minutes, yes. But that's not the part that takes all the time.

The part that takes time is humans playtesting it and then deciding what the tweaks should be, if any, making the tweaks, then going back and playtesting it again to see if the tweaks made the game more fun.