In my experience, they just pretend it doesn't exist and remove it from the "plan". Where I see this the most is in figuring out how things work. For some reason, non-programmers have it in their head that once you've finished college, you ought to know (even though they don't) how everything that goes on a computer works instantaneously. We're going to use Apache Kafka! Ok, need to spend some time figuring out how that works. Um, ok, that should take like an hour, right? Because there's no time in the budget to learn anything. If you push back, they say, "ok, never use anything new that you have to learn. Only use the things you already know". That doesn't work either, because the things that we have to integrate with don't work with the old things. Or, "you don't need to know how that works, Bob knows how that works, just ask him. But don't let it take any time, because there's no time in the plan for that" And Bob is sick of all your questions that you'd know the answer to if you'd just spent a two or three days reading the online documentation and installing a test system and testing it out.
Most of that will be doing spikes: http://agiledictionary.com/209/spike/
It's one the most underappreciated agile concepts. I wish people would shut up about standups once in a while and give more thought to doing spikes.
Spending 30 minutes researching four different libraries to see which one to would be a spike, for instance, but it doesn't necessarily involve building a proof of concept.