As people who know me professionally and/or read my blog know well, I have been a (raucously) loud evangelist for test-first development (TDD, BDD, Scrum, whatever your flavour) for years now. If I write even an exploratory bit of code and don't have tests in place first, I get very uncomfortable. As complexity increases, without tests (preferably automated, repeatable tests), I argue that I simply can't know what's really going on, because I can't prove it.
A major corollary to this is test coverage reporting. If I can't see what's been tested and what hasn't, then in a very real sense nothing has been, since I can't document/prove what has been and what hasn't. And the better (more productive) teams I've worked in have established, and regularly hit, coverage testing better than 95%, with 100% being a common (and commonly attained) goal. (Edit: Note that this is for C0 and C1 test coverage; tools that cover C3 and C4 are rare to nonexistent in most languages, such as Ruby.)
As you may also know, I've been getting (back) into Ruby development, using Rails 3 on Ruby 1.9. Ruby's long-time de facto standard coverage tool for many years was rcov, which generally worked well. However, Rob Sanheim has stated that "RCov does not, and will not support C based Ruby 1.9.x implementations due to significant changes between 1.8 and 1.9.". He recommends either SimpleCov or CoverMe. Ripping out RCov and replacing it with CoverMeSimpleCov on a test project took all of five minutes and left me with attractive, functional, (so far apparently) quite accurate reports.
One of the basic principles of agile development is that the team must actively embrace constructive change as their project evolves. It's often easy for harried, hurried people to forget that that applies to their tools as much as it does to what they produce using those tools.
Just a thought as my evening winds down.