We have developed a method for accelerating equilibration in numerical simulations of glassy models, using the Frenkel‐Kontorova model for developing and testing the algorithm. This simple model is like a glass in that relaxation times diverge rapidly as the critical temperature is reached, so that no matter how slowly it is cooled, the system eventually gets stuck in some random metastable configuration. By adding ‘‘numerical enzymes,’’ or long‐range Monte Carlo moves which precisely eliminate certain barriers to relaxation, we can equilibrate rapidly to significantly lower temperatures. Our numerical method for developing these enzymes is a learning algorithm motivated by Darwinian evolution.