Most of the times you'd be right. But some of the times you'd be devastatingly wrong. You always need to check your assumptions about runtime, expected behavior and so on and there is only so much modeling you can do in your head. At some point - if the scope is too large to fit in your head at once - you will need to resort to divide and conquer, implement the part that you understand, regroup, then do the rest. And before you know it halfway into doing 'the rest' you will realize a better way to approach the problem.