In Carmack's case it was "lock yourself in a room for 14 hours a day coding" and get yourself a hype man (and artist/programmer) like Romero. More seriously, find a software niche that hasn't been widely marketed yet. You don't need to invent anything, just find something that's being done but hasn't been done good enough to reach a wide audience. Network with people that have a similar interest and form a company once you've got an idea worth pursuing.