I will discuss the problem of building machine learning agents that help other agents reach their goals. This is a core problem underlying AI-assisted decision making, design and modelling which is FCAI’s joint methodological goal, and where the other agent usually is the human decision maker, designer or modeller. FCAI applies the methods in virtual laboratories across R&D fields. I will start with examples on interactive knowledge elicitation, and continue to AI-assisted design, user modelling for sequential machine teaching, and POMDPs for centaurs.