I've made this mistake in the past, but I don't any more. The difference is that rather than closing my feedback loop over the code/architecture itself (so the

I've made this mistake in the past, but I don't any more. The difference is that rather than closing my feedback loop over the code/architecture itself (so the cycle of working is start feature => code => look at code => code, which obviously never leads to starting new features), I close the loop over the experience (becoming start feature => code => use/test the feature => start new feature). This way your focus is on the product and your working cycle involves incrementally adding features, not navel-gazing at your code. So basically, bring yourself out of the code and actually try the product whenever a feature is done; context-switch from coder to user.