The development of individuals' digital skills has received much attention as a remedy for digital inequality. Although some researchers favor courses and guided learning for skills development, others propose learning by trial-and-error. Unfortunately, studies examining the value of the so-called "do-it-yourself approach" for the development of digital skills remain lacking. One difficulty lies in the vicious circle of lack of skill leading to infrequent Internet usage and vice versa, which limits the value of cross-sectional data for assessing the impact of this approach. We present longitudinal data on a random sample of Internet users in a Dutch city, which show that more frequent Internet use leads to more digital skills, but not the other way around. However, contrary to expectations about the potential of trial-and-error learning to reduce inequality, results also suggests that this approach is not always more beneficial to the "have-little" as compared to the "have-more." The only inequality-reducing effect of this approach is that that older users profit more from it than younger users do.