Reform the grant application process

The way research grants are evaluated and awarded makes bureaucrats out of scientist.

That's the bottom line as presented in a perspective article in PLoS Biology.
Real Lives and White Lies in the Funding of Scientific Research:
The granting system turns young scientists into bureaucrats and then betrays them

Here's the cause of the problem (wrong selection pressure):

The peculiar demands of our granting system have favoured an upper class of skilled scientists who know how to raise money for a big group [3]. They have mastered a glass bead game that rewards not only quality and honesty, but also salesmanship and networking. A large group is the secret because applications are currently judged in a way that makes it almost immaterial how many of that group fail, so long as two or three do well. Data from these successful underlings can be cleverly packaged to produce a flow of papers—essential to generate an overlapping portfolio of grants to avoid gaps in funding.

And here's the consequence (better grant-writers, less innovative research):

Thus, large groups can appear effective even when they are neither efficient nor innovative. Also, large groups breed a surplus of PhD students and postdocs that flood the market; many boost the careers of their supervisors while their own plans to continue in research are doomed from the outset. The system also helps larger groups outcompete smaller groups, like those headed by younger scientists such as K. It is no wonder that the average age of grant recipients continues to rise [4]. Even worse, sustained success is most likely when risky and original topics are avoided and projects tailored to fit prevailing fashions—a fact that sticks a knife into the back of true research [5]. As Sydney Brenner has said, “Innovation comes only from an assault on the unknown” [6].

And, finally, the solution:

Drastic simplification of this grant-writing process would help scientists return to the business of doing science. Grant applications should be much shorter, and if so, scientists would spend less time writing them and less time reviewing other applications.

(...)

The solution is to allow, say, only three to five of the best papers of the group from over the previous five years to be offered for assessment (as Howard Hughes Medical Institute already do in the US). The evaluation of these papers should be corrected for the size of the group, i.e., productivity would be rated per person, not per group. If these reforms were enacted, the pressure to rush out many papers would be replaced by pressure to complete projects that report stories of value and present them well.

Asses grant applications on previous work, rather than future work. Previous work documents that the P.I. can get research done, and what the quality of that research is. This would also, crucially, free up time for researchers to do research, rather than writing grant applications.

Seems like a no-brainer to me.