They have a nice model. The program managers (those who set research directions) are fixed-term IIRC, so they have very little incentive to politic, and instead can focus on seeing their pet program through to completion in the given time / funding.
One of their main "jobs" (aside from above), is to be an advocate for emerging tech among the military and civilian industrial partners to make sure it actually "becomes something".
To work at DARPA in this capacity usually requires a lot of successful R&D program management experience and some sense of grand vision. To work in a lesser, administrative capacity, including interacting with research partners, requires just a tad less, but on the average they've been exceptional, intelligent, and responsive partners for all the work we've done for them.