There are a bunch of software environments in which you can make instruments. Max/MSP, Kyma, AudioMulch, Reaktor, Vaz Modular, Tassman, Plogue Bidule ... and that's just the graphical ones I can think of, leaving out SuperCollider, ChucK and so on. How many times have you read that with tool X you are "limited only by your own imagination?" But not everyone has the ability, or the time, to think up a new instrument and then spend a year working out the details. For players and composers, limitations are essential. So where are the instruments themselves?
With computer audio, there are so many possibilities available that I think the idea of an instrument is getting neglected. A traditional musical instrument usually has a small number of controls that lead to many possibilities, because you can physically interact with it in different and subtle ways. Take a single string and stretch it over a box with a hole in it. You just made an instrument that you could spend years practicing and getting better at. There aren't many parts, but the variety of physical interactions they offer gives you lots of sonic possibilities.
Contrast that with a computer, which has billions of parts that can go into more states than the universe has particles, most of which make no sounds at all. It's hard to configure these systems to make sound dependably---recall only ten years or so ago, when just a handful of brave musicians were willing to rely on computers for live performance, and crashes were pretty common.
Once you do make sound with a computer, you have entered an exciting world where it's arguably possible to make any sound we are capable of hearing. Where to begin? It's natural that the environments people have built for harnessing all this power look like the technology they are based on: component-based systems where simple building blocks are connected to make larger components, and so on. This kind of approach has been used by every music synthesis language or environment that has ever been made, as far as I am aware. It makes sense when you don't have much computing power, or you do want to express musical ideas in terms of algorithms.
But physical instruments are not algorithms. We don't send commands to them, we play them, using vibrating surfaces to exchange information with them more subtly than symbols can. With the amount of computing power available today, it's possible to make fantastic new computer instruments that are more like physical objects than programs, expressive enough to be worth learning.
This has been a Madrona Labs mini-essay. Thanks to M-Goldie on Gearslutz.com for asking the question.