We contribute a detailed investigation of AR-enhanced widgets for smartphones, where AR technology is not only used to offload widgets from the phone to the air around it, but to give users more control on input precision as well. Such widgets have the obvious benefit of freeing up screen real-estate on the phone, but their other potential benefits remain largely theoretical. Their limitations are not well understood, most particularly in terms of input performance. We compare different AR-enhanced widget designs against their state-of-the-art touch-only counterparts with a series of exploratory studies in which participants had to perform three tasks: command trigger, parameter value adjustment, and precise 2D selection. We then derive guidelines from our empirical observations.