The rate of job-to-job transitions is twice as large today as the rate at which workers move from employment to unemployment. I demonstrate that, under plausible specifications, the basic job-ladder model --- the workhorse model of the literature on on-the-job search --- has no chance of matching the extent of job-to-job transitions. Moreover, it cannot account for the low search effort exerted by most employed workers and the observation that on-the-job search is a means to ``escape'' unemployment: it is undertaken exactly by those workers who are facing the threat of becoming unemployed. I develop an alternative theoretical framework that can quantitatively match salient features of job-to-job transitions. The model incorporates a stochastic process that causes the value of a job to the worker to decrease at times, predicting that workers with a lower job value have a higher probability of entering unemployment. This natural feature is not in standard models. A second important element of the model is endogenous search effort, explaining the low search effort exerted by many employed workers and the correlation between search effort and the probability of becoming unemployed observed in the data. Calculating the equilibrium of the model shows that it can successfully account for the stylized facts on job-to-job transitions. I also demonstrate that the model can account for observed differences in the extent of job-to-job transitions across demographic groups