The problems we now see with the global markets and currency exchange system result from trade-offs made to solve historical problems.
One of the major problems is human neurophysiology. The median market participant cannot keep track of an entire matrix of the relative value of goods and services, availability of supply, and such, after the number of distinct market options pass a certain threshold. That table grows at O(n^2). So the market chose the m commodity goods with the most universally stable demand to act as currency, and the table then grows at only O(m^2 + n), which is much more manageable.
So if you brought butter and milk to the market, and took away flour and eggs, instead of needing to know 16 values, you traded your goods for coins and your coins for goods, and only needed to know a minimum of 1 values for each of the 4 goods, plus all the exchange rates between coins.
We don't actually need to do that any more. Computers are great at remembering huge quantities of numbers, and algorithmically traversing graphs. In theory, you could add a node to the market graph that describes the good or service that you offer, add a node that describes the good or service that you want, and a computer could grind out a solution that gives everyone in the trade chain as much of what they want as possible, for as little as they can spare of what they offer.
The default system as it is now is that everyone in the US always has a want open for USD, with no numeric cap, and usually also has an offer open for as many USD as they can afford to spend. So when you offer something, someone will pay USD for it. When you want something, you give USD for it. You don't have much computing power to waste on graph traversal, and certainly aren't trying to maximize utility for anyone other than yourself. So you don't automatically get the best possible deal.
When you rely on a computer network to do all the heavy lifting, you don't need your default settlement currency to be the same kind that the local ATMs spit out. You could decide that you would accept unlimited quantities of USD, EUR, GBP, .9167 Au, .4000 ethanol, bitcoin, litecoin, dogecoin, or etherium. Just like that, you're writing code for a pizzeria, and some mason jars filled with 80-proof appear on your back porch. You have no idea how the pizzeria connects with the distiller. You can only assume that the distiller wanted some pizza, and the pizzeria wanted software-authoring services rather than illegal liquor. Since your code is worth more than the distiller's pizza, the pizzeria also gives you the barbecue chicken and sweet onion pizza you said you wanted and some USD. Some bitcoin appears in your wallet, probably from someone else who wanted pizza, but couldn't get it directly from the pizzeria because they didn't want to accept bitcoin. No natural human trade network could ever routinely clear multiparty trades that complicated. But doing huge quantities of tedious and complex math is a cinch with computers.
As soon as enough people commit to accepting alternate currencies, or bartered goods and services, a whole lot of trades that had previously been stalled, waiting for a payment solution, start clearing.
The existing currencies have always been used to reduce the otherwise burdensome cost of trade. If you annihilate that cost with automation, you don't need the common currency any more. That pizzeria can start to pay its suppliers, contractors, and employees with pizza IOUs, and those circulate as another form of currency, backed by both the promised producer of the pizzas and those who want to eat them. Without the former, the IOUs could not exist, and without the latter, they would be worthless.