Teaching computer-assisted translation (CAT) can require providing students with direct instructions on how to operate specific CAT tools. As opposed to subjects like translation theories, the teaching of CAT is often expected not only to instigate critical reflection but also to develop students' instrumental software skills. Striking a balance between these two expectations is not always straightforward, however. This article presents an exploratory investigation that attempted to reduce the use of direct instructions in CAT teaching. We used different combinations of eye tracking, keylogging and screen recordings to examine the performance of students with no CAT experience in three autonomous learning tasks. We found that most students were able to complete the tasks with reasonable success despite the lack of direct instructions. We draw on the education literature to discuss the results and call for further research on the role of tool-specific instructions in the teaching of CAT.