It seems to be quite popular for folks to vent about the AWS SA interview process on GD after the fact, and I did find it difficult and awkward, but not in the ways I was expecting. I only made it to the initial 90 minute phone screen, and did not get selected for the full onsite loop.
First off, the good things:
1. Recruitment team were really helpful and professional
2. Overall it is a fair process, and definitely gives you an insight in to how the group operates.
3. Reinforced my belief that you would be joining a high performing team.
Not-so-good things:
1. Phone screen with team is done on an awful in-house tele-conferencing line with poor audio. I couldn't always hear what they were saying, and vice versa. This definitely a barrier to proper communication and not in the candidate's favor. Make sure you're in a quiet space with excellent phone hardware :-)
2. I spent lot of hours/days/weeks prepping for the interview, but I spent far too much time on organizing my answers to the behavioral dimension of the expected question set ("Tell me about a time" etc) and not enough on the base AWS capabilities. For the initial tech screen with the hiring team, it is very important to be able to virtually whiteboard (ie, describe) how to build a scalable and secure architecture on AWS. This is on me, my judgement was way off here, and I should have spent more time rehearsing this aspect, and left the 14 principles to the onsite loop (I did get asked 1 or 2 behavioral Q's on my phone screen, but they were almost after-thoughts).
3. There are some open-ended questions centered on industry buzz-words that are, perhaps deliberately, ambiguous, but when pressed for specifics, it seemed the interviewer could not provide actual details. I am not totally sure they always knew exactly what they were asking, and were working from a script. For example, there is a fair few questions about distributed architecture with no actual specifics around scale or business problem. Whether a computing system can be labelled "distributed" is very much in the eye of the beholder, in my experience. Scalability or elasticity does not imply true distribution.