Solving the 'right' engineering problems

In engineering, it's easy to find problems. The code is never 100% perfect. The product never has all the features. But which problems are worth solving?