Abstract: Magnetic reconnection events are transient phenomena: a relatively quiescent plasma suddenly transitions to a reconnecting state where large amounts of magnetic energy are released. This sharp transition is known as the onset of magnetic reconnection. Analytical and numerical studies of reconnection often focus on assumed post-onset configurations and plasma parameters, by-passing not only the onset transition, but also the question of how such configurations came to be — and, perhaps more pertinently, of whether they are even physically realizable. For example, we now know that large-scale Sweet-Parker-like current sheets cannot form dynamically because they are subject to the super-Alfvénic plasmoid instability. Similar issues arise in collisionless plasmas, where the natural evolution of a given system leading to the formation of current sheets can trigger a variety of instabilities before such sheets reach electron scales. These notions imply that the onset of reconnection can be obtained by following the current sheet formation process and its transition from a stable to an unstable system. Moreover, I will argue that, because instabilities are inevitably triggered during the current sheet formation process, the (nonlinear) reconnection stage cannot be trivially decoupled from the current sheet formation stage; for example, the parameters characterizing the reconnection event may depend critically on the formation process and the instabilities that arise then. I will illustrate these issues both in the MHD and the kinetic regimes.