The challenge of building large-scale complex systems often gets caught between purist visions that never get off the ground and seemingly pragmatic random-walk tinkering that slowly grinds to a halt via diminishing returns.