Automated Candy Factory

So about a week or so ago, I was watching an episode of one of those factory shows, where they detail the start-to-finish manufacturing of something. The episode was about candy. How to make jellybeans, jawbreakers, chewing gum, and things like that. I noticed something very strange: Everything was only semi-automated.
Yes, much of it was automated, using conveyor belts, rolling drums, mixing vats and so on. But so much more was done by hand. Sugar and flavoring could have been added via a hose, as they were introduced in liquid form, but were instead ladled in by hand. Transferring from mixing vats or rolling drums could have been done by upturning the vat onto a conveyor belt...but the candies were moved by hand instead. Why?
Okay, sometimes people break automation steps as a quality control measure, on the grounds that a human being moving candies can notice if one is misshapen, unevenly colored, or in some other way defective. On the other, shovelling candy from a mixer to a conveyor belt seems like a really crappy job to have, and the pay can't be that great. Also, the candy workers have to put a lot of effort into remaining sterile. People are going to eat these candies, and people do occasionally have to touch really gross things. I'm sure the workers do great at that, as food poisoning from candy has not happened (to my knowledge) in the entire history of the united states, but with automated equipment it would not be possible at all, as automated equipment would touch nothing but candy.
If one did create an automated candy factory, it would operate 24/7 for cheap. And clean. And reconfigurable by computer. (Well, okay, not by default, but the way I'm imagining it, it would be entirely computer-operated and controlled.) With a computer controlling everything, and able to arbitrarily rearrange the factory, the factory could, with its excess capacity, do some experimentation. Try mixing flavors, or varieties, or try to come up with entirely new candy varieties. Ones no human had ever thought of. In fact, let's apply genetic programming here. Experimental candies are tested with focus groups, and their results quantified. Candies that test well feature heavily in further experiments. Candies that test poorly are avoided.
For randomization, programmers would have a Malbolge-like interface. (Malbolge is a programming language designed specifically to frustrate programmers. Their are multiple steps where the code is encrypted in a way that doesn't make intuitive sense, making all its programs difficult to understand in any way. A finished program resembles a chunk of line noise.) Their flailing attempts to design candies are the seed genetics, which get refined from there.