How can we get better at starting smaller - i.e. with prototypes - rather than over-investing in scalable solutions by default?