Contract work on modular functions, not a monolithic code base. Spawn a child, apply contractor's code, watch what happens. It would be a mixture of Genetic programming (fitness tests) and intelligent design :)
Up to a point; often the module boundaries are wrong (I've worked in a company that failed mostly for this reason - team A worked hard on module A, team B worked hard on module B, but neither was able to challenge the insanity of calls that had to ping-pong back and forth between A and B several times over to get a result). And you can't touch the genetic programming structure itself; the program can optimize itself, but it can't optimize its optimizer, and ultimately it may be outcompeted by programs with better optimizers.
In traditional capitalism the buck stops with the board, and the shareholders will fire them if they're losing too much money (something that's become more important in recent years, even as more and more day-to-day stock trading is algorithmic). Who takes that role in the case of an autonomous algorithm - would the algorithm itself be a director?
