Most of it is not an individual experience or 'event', just bad design with bad results. I'll try to describe some global ones:
One of the most bizarre things is the crazy bad resource hierarchy. There are multiple overlapping and incompatible ones. Resources, networks, storage, IAM, billing and org, none of it in a single universal hierarchy. It seems to mirror the idiosyncrasies of legacy enterprise organisations with their fiefdoms, instead of a cloud.
The next useless thing is how you just cannot use what you need when you need it in whatever way you want it. Almost all services are hyper-segmented requiring various premium tiers instead of them being universally available.
I get it, it's a great way to bundle things people don't want and extract as much money out of them, but that only really works if people have no alternative. And those two form the bad architecture/bad technology trifecta with this third one: a lot of services, maybe most of them, seem like some sort of 2005 model where a resource is backed by nothing more than some random managed VM in the backend, with all the problems (failure modes, inconsistent behaviour etc) that come with that model.
Perhaps the reason for those things is simple: Microsoft wanted a way to extract more money from their customers and lock them in even more. Moving workloads to Azure meant something different for them than it did for the rest of the world: you used to have a lage army of common windows sysadmin jobs where there was a lot of local control and local management loops, but when you move that to a common template in someone else's datacenter (Azure, essentially) you can ditch most of those loops and people. Granted, they created those local controls/loops themselves to get a school-to-work microsoft client pipeline (same as say, Cisco or oracle), but I doubt there is any new markets to cater to in that way. Since people tend to be the most expensive and most risky part of a business, being able to take more of them out of the loop, making more of them optional or making them more expendable/remote is a (short-term) positive thing in the spreadsheets of most MBAs, which is who most large companies cater to after all. This did of course backfire and we now have the same quantity of jobs but instead of sysadmin you get 'azure engineer' which is more of a crossover between operational helpdesk and technical application manager. But everyone wins: during your exodus you can sell it as modernisation, when you remove that on-prem burden you can shift your CAPEX and OPEX around, your quarter looks better when you can reduce headcount, and once your bonus is in, you can put some job postings out for the people you are now missing.
Technology-wise, the only thing that really changed was the ways in which people could cut corners. Some corners are pre-cut, while others are uncuttable. API-wise, it's a crapshoot, a turd attempted to be polished by a webui that hides the maelstrom of questionable residue below the surface.