still need people with operational knowledge
Because 95%+ of developers don't know how to operate shit. I speak as someone who has worked closely with developers(including 2 companies that used AWS for some time) for the past 16 years at a half dozen different companies(mostly startups).
Most of them are completely clueless when it comes to operating stuff. Many of them don't even care, they don't WANT to know they just want to code. Code quality is quite bad and generally is trending towards worse rather than better.
I remember one of my former co-workers told me a little thing that happened at his company this was probably 5 or 6 years ago, they were in amazon and the developer asked "can't you just turn on the "auto scaling" feature?" I could understand that coming from a non technical person but a developer? I don't have words.
A lot of orgs that use public clouds, especially IaaS (SaaS does make sense in a lot of cases), don't know any better. I mean they don't know that it is unusual to spend six figures a month on services. They don't know that they can cut their spend significantly by doing things in house. Of course you need the talent to do so.
One excuse I heard for using public cloud(the company wasn't using Amazon, but another smaller enterprise player I forgot the name, not one of the obvious ones). This is a billion dollar corp, spending more than $600,000/month with public cloud. They had regular performance problems and outages. But the management didn't care, they didn't want to have to "deal with servers and vendors". (even though they had to deal with their cloud vendor all the time). One of my friends worked there and proposed a plan to bring it in house with a 4 month ROI, management didn't care.
Last company I was at was paying at the peak over $400,000/mo to amazon, company imploded not long after I left. My current org hired me to move them out of amazon years ago, they launched their app and of course the costs exploded as they tend to do with cloud services, towards the end of our time in that cloud we were well over $100,000/mo, and I'd argue we have grown 8 fold or more since that time (with that growth all of the gear fits in roughly 4 cabinets). The savings were obviously huge, but savings aside the higher availability, better performance, more control, is not as easy to quantify as the raw $ savings from purchasing/etc alone.
My current engineering management had cloud experience in the past(Joyent I think), with similar results - spending upwards of $500,000/mo on services resulting in a plan to bring it all in house(and they did), so fortunately I haven't had to fight my butt fight in some time. Don't get me wrong though the list of SaaS services we use is long as hell, dozens of them(even some for Ops including stuff from Dyn, Neustar, LogicMonitor, Pagerduty, Duo security etc).
I'll also say many ops people are shit too, run of the mill IT staff have been bad forever probably (few exceptions I'm sure).
Operating in a public cloud, especially one as limiting as AWS is requires talent(to do it right), I argue more talent than is needed to do things in house. There are a lot of "features" that you don't realize you get up front.
It certainly can work, if you put enough effort into it, or if you just live with the unreliable nature of the platform(along with costs etc).
Amazon tries to recruit me at least twice a year, along with netflix and I don't try to keep track how many service providers and cloud companies and other companies etc. I'm happy where I am at though.
Looking forward to tonight, HooterPalooza 2016, hooters bikini contest in Fremont, CA doors open at 5PM.