Make each feature work hard to be implemented. Make each feature prove itself and show that it's a survivor. It's like "Fight Club". You should only consider features if they're willing to stand on the porch for three days waiting to be let in.

That's why you start with no. Every feature request that comes in to us — or from us — meets a no. We listen but don't act. The initial response is "not now". If a request for a feature keeps coming back, that's when we know it's time to take a deeper look. Then, and only then, do we start considering the feature for real.

Auteur: 37 Signals