[zimbatm]

I think it's a natural part of the learning process. Either abstract away (usually with leaky abstractions) or simplify by rewriting (usually with the same result once the complexity of the domain is factored in). In some cases it produces something better but usually not on the first iteration.