It depends. At low to medium scale its a case of keeping your eye on things and let the system sort itself out.

At high scale you do want to get under the hood and start playing. However we should always strive for automation. If you can determine the decision making points and processes a DBA uses to decide a course of action then you can automate that.

Remember when clustering involved a lot of hacking the registry and other under the hood voodoo? Well it doesn't anymore and some of the things we have to do as configuration options today will vanish tomorrow.

Remember DBCC PINTABLE? Gone now because memory management is so much better.