Ultimately I think any truly autonomous agent would be out-competed by one that has a human in the loop modifying the code to adapt to the market (e.g. customer development). But it's a cool concept.
I think you'd have to have some pretty cutting edge code generation genetic algorithms to have anything like this outperform a human, unless I am mistaken about the state of genetic code generation.
Genetic algorithms excel at a particular set of problems. Those problems are pretty general in scope, and are often tricky to solve by humans that don't have the requisite intuition or time to explore the problem space more thoroughly.
You can't solve chess with a genetic algorithm, at least not to my knowledge, but you could sure as heck make a challenging checkers player with one.
Real world businesses require the integration of many knowledge domains in order to succeed. The specialized niches where genetic algorithms excel are at best a small subset.
Some similar musings from history: http://www.research.ibm.com/massive/bump.html http://www.cap-lore.com/Agorics/Library/agoricpapers.html