You're never really starting from nothing. We've historically been taught that to start a project you first pick a subject, but you can't really pick a subject if you haven't already put thought into that subject, at least to know that it is a subject that can be selected. This begs the question where do we really start? You start with research, research which is completely unguided by any preconceived notion of an objective, it is guided by exploration and learning. From this you identify a problem you want to work on.
When building a startup from Zero-To-One, it's essential to create an environment that allows the team unfettered exploration. This doesn't mean you have no direction, but if you don't have unfettered access to customers and the ability to see where they bring you then you're hindered by the preconceived notion of a plan which is selected arbitrarily.
Once you've laid roots in an idea through unfettered exploration and learning you can follow Eric Ries advice in the Lean Startup: follow the scientific method starting with your hypothesis and then Build-Measure-Learn. Build the smallest possible version of your idea needed to achieve Validated Learning about your hypothesis.