Agent-based modeling is a promising tool for assisting decision-making in, and reasoning about, complex systems. However, using these models in practice is rife with challenges: their complexity can make them difficult to analyze mathematically, their sheer scale can make numerically simulating them computationally expensive, and their granular and high-fidelity nature can make them difficult to interpret and explain. Our research group at Oxford has been developing tools and methods that are grounded in modern computational statistics and machine learning, and whose purpose is to begin addressing some of these challenges. My talk will provide an overview of the current state of this body of work and the different themes involved, which include parameter estimation, population synthesis, causality, and building and exploiting gradient information in automatically differentiable agent-based models.