The second part of the question is "opinion based" and out of scope for this site..
But, simulation is not something meant to replace real world testing. It's another tool in our development toolbox. It has some advantages but also some pitfalls. The simulation is as good as the model you use.
One key aspect of simulation is the ability to "mesure" quantities that are not mesurable in the real world. For instance internal timing inside a chip itself. One can't have access easily to the chip internals.
Another aspect is the speed at which you can adjust and try new things, without creating a new prototype at every step.
As I said it's just a tool. Most of the time, some part of the design are simulated. When others are not.
But every simulation tool user knows (or should know) that simulation models are not perfect and thus the simulation has some limitations. "It works in simulation" doesn't mean "it will work in the real world" ! At the end, only real world testing, at different temperature, load, batches of critical components, etc is able to validate the design against its spec.
A simulation of an OPAMP negative feedback might tell you that you have a phase margin of 60°. Nice, but that simulation doesn't tell you what the real phase margin will be. The only thing you can conclude is that your system is very likely to be stable because you have a big simulation margin and if the model is not very poor, it should agree with the real thing somehow. I am not saying that all OPAMP simulations are not accurate, it was just an example. You may find extremely good models that accounts for temperature and even some parameter tolerancing that will results of excellent results.
But if you are simulating "basic stuff" : resistors, inductors, MOSFETS, BJT etc, the probability of having a correct answer that closely match the real world is very high, and you may not have to prototype that part of the circuit.
The point is to choose the right tool for the right job. As you won't use the general relativity to solve for the motion of an apple falling down a tree. Even if you know that Newton's laws of motions are a just a special case that works well only in specific conditions and if theses specific conditions are met, go for Newton ! It's much easier and faster.
Same for electronics simulations, you are designing 60 GHz filters, don't use a spice simulation using the internal R L and C models, you won't go very far. Here the right tool would be an RF aware simulation package. But simulating a MOSFET for a LED driver at 100KHz at ambient temperature, It won't make any sense to use Agilent ADS for that task. Spice is perfectly suited to that task and may allow you to pinpoint problem such turn-on time out of specs...