- EC2, including scaling groups (but crucially do not attempt to use them for auto-scaling)
- S3
- Route53
- RDS
- Redshift
- Maybe a few others I've forgotten or haven't used
They all have their flaws, but if you stay on the beaten path, they basically work most of the time.
Then there is the other AWS. It consists of all the other high level services you can pick from that enormous drop-down. I haven't tried many, but from my little experience and loaaaads of anecdotes from other people, _that_ AWS is largely unreliable rubbish.
That is to say, stick to the extremely common low-level or otherwise foundational stuff unless you love pain and disappointment.
But the thrust of your point remains: when you get into the nonsense factory, when you're trying to bolt together the higher level integrated services, it becomes an exercise in self-inflicted stochastic violence.
The true AWS advantage is that the boss can't get fired for choosing it so the individual contributors already have spending authority inside its walls. Unfortunately, that's a trump card.