I certainly do not expect GM mode to be anywhere near as flexible as a real P&P game.
However, I feel the most important part of the experience of having a GM is the flexibility of the storytelling. Flexibility of actions is less important. That is, I feel it's not important that GMs are able to improvise a mechanic like climbing a wall or jumping through a window. But if it's appropriate for a character to react in a certain way during a conversation, then the role of the GM would ideally be to facilitate that conversation, and enable the story that unfolds.
Procedural generation is something above and beyond that, however, and it acknowledges that GMs are likely to be very much amateurs. I expect we would have a ratio of 1 GM for up to 4 players, so potential GMs (like myself) lack the talent or skill we might expect from "modders" who are releasing their mods to much larger audiences.
I'll leave it up to the developers to work out what is practical from the perspective of budget and development time. Procedural generation doesn't have to live up to the expectation of being as good as handcrafted maps. Even if it only gets you halfway there, then that's half of the work done already. The ideal is to minimise the amount of micromanagement a GM has to do to get something functional, while giving them as much control over the results as they need.
There are modest versions of procedural generation. One is improving the terrain "painting". For example, I could paint a forest of trees over a particular area, and the engine would select individual (and varied) trees to place in that zone, instead of forcing me to individually select and place every single tree. The game could generate suggested names for NPCs, stock for merchants, etc. The game already has random loot, and I expect we'll be able to generate random encounters. These are all small parts of this picture that I don't always want to micromanage.
Procedural generation is something that works best when it's part of the design from the beginning, because if the tool is good, then Larian would be able to use it themselves when making the game. Procedural generation is already part of the game. I just really want to emphasise my view that "more is better".
I do feel that procedural generation has a lot of potential to be a key component of the GM experience.