We reflect upon a paper we wrote that was published in 1995 (20 years ago). We outline the motivation for the work and situate it in the state of the art at that time. We suggest that a key contribution was to highlight the need to provide support for learners who reason with external representations. The support must be flexible enough to accommodate wide individual differences in representational knowledge. In 1995 we saw AIED systems as potentially having several roles; one key role was to enable the learner to move easily from one external representation (ER) to a different one (switching ERs). Other necessary roles included: to seek to use AI to keep several forms of ER in synchrony as the student works on one of them; to guide the student to one ER form (out of several) most likely to be of use; and to help the student use an ER effectively. We outline where our work seems to have had influence and we review progress in the field since 1995. In our conclusion we identify future research agendas that emerge from recent developments in cognitive science.