I think there are a lot of factors at play in what you're describing. Sorry for the wall of text, tl;dr: skip to the last section about startups undercutting behemoths.
> My first impressions is that everyone knows only quite narrow field of the project
When you reach an employee count that can sustain it, this is the "optimal" way to organise a company. Division and specialisation of labour is how Big Corps operate. At that size, one person can't know almost-everything like they can in a startup. They most likely don't even know everything about their own product, depending on its size.
> Next thing that really struck me was how hard is to just approach someone and get to know him
This is both a function of the people themselves, and the amount of stress and/or deadline pressure they're under. In both cases you can imagine people not having time to get to know someone new, particularly if they're not on the same team. This is not an insult, it's just a function of prioritising who and what to interact with in a company of ~300K.
> As there are people sitting next to each other that have way different work.
This also exacerbates the above and is one of the key reasons I loathe hot-desking. If you're going to colocate people you should colocate people on the same team where possible. That is a hill I will die on.
> it is hard to approach someone since most of the people spend large amount of time on online calls
Also common at Big Corps, but there's also an explosion of this since covid / remote work normalisation, etc.
> Therefore you really never know if you can approach someone and don't look like an disturbing idiot ... I mean the people are actually pretty nice, they are not stupid at all, but everything just seems so dull.
Assume everyone is friendly. "Excuse me, sorry to disturb you, [I'm new here] ..." is a great opener if you're worried about this. Their reaction will tell you whether your initial assumption is valid at that time in that context.
> And I have feeling like it really does not matter if you are really good or under average.
Different companies handle performance very differently. My advice is to not tie your worth or goals _specifically_ to the company's performance measurements, but look at your own personal growth as the yardstick for how satisfied you are with your role and progress. Then only interact with the company's performance model where necessary to succeed within the box the company is imposing.
> I feel like we are dealing with not that much of complicated technical products, but the complexity of work is created by all the processes (SAP, etc.). And I have feeling like it really does not matter if you are really good or under average
Complexity of technical products obviously varies wildly within a company and across companies, but in general yes: process is a defining hallmark of a Big Corp. Process is essential for replicating the success they've achieved and maintaining the redundancy needed to take risks and fail.
> and suddenly can understand why startups can undercut such behemoths.
I think this half of the key observation. The other half is: every bet a startup makes runs the risk of killing the company. A Big Corp can make 100 bets and fail every single one of them, and still operate successfully as a company.
But that doesn't mean they're any more likely to win any individual bet -- innovation is hard -- they can just make a ton more of them without going bankrupt. They can only do that because they've built an enormous, redundant, safe, "dull" system that has the best chance to grow over time with their existing successes.
The only way a startup can compete with this is to find the one thing that a Big Corp can't/won't do, hasn't noticed, won't admint, can't move fast enough to capture (e.g. new technologies), etc, etc, and successfully execute on it, fast, with an entire company of like, 10 people.
Scary if you're betting with peoples' livelihoods, intense, and requires an enormous breadth of knowledge across 10 people to operate in all kinds of roles. But certainly not dull.