As a side (real world) example case of the huge gulf between routing and "intelligent" routing, almost two decades ago, the "shop" I was working in moved to a SP WAN (international) cloud, where routing interchange was BGP.
One day, one of their interior nodes decided to black hole transit traffic. From a BGP perspective, to us, nothing visible (for whatever reasons, issue wasn't visible to SP [a tier one SP], and took hours to track down and fix).
(What was visible to us, a few specific sites were experiencing normal or total network failure between just those sites. If you pinged between those sites you would see between 50% to 75% drops.)
Network manager came to me, as we had SP redundancy, is there any way we can avoid a SP issue being so impactful?
I said, possibly yes. At that time I had been using the (then new) OER (Optimized Edge Routing) technology on our Internet facing routers, and thought it should be able to work its magic on our SP cloud. (Which it did, very well.)
What was the basis of its magic? It passively (optionally actively) monitored end-to-end flow performance across multiple paths and it worked to achieve optimal performance, by rerouting flows, often within seconds.
So, in the case of some flows being unable to transit end-to-end across one SP, it would move the flows to the other SP.
When I started to use this (proprietary) technology on our WAN, the only complaint was from our network monitoring group which no longer saw any WAN performance issues.
OER is likely the distant forerunner of much of SD-WAN.
I could go on and on about the wonders of OER (and its immediate successor, 1st gen PfR [Performance Routing]), but what they could do, in near real time, was anywhere impractical to impossible using any dynamic routing protocol at its best.
BTW, another possibly con of using such advanced technology, that can be overlooked or under estimatd, you need higher skill set to optimally use such technology.
