A model for the joint evolution of opinions and how much the agents trust each other is presented. The model is built using the framework of the Continuous Opinions and Discrete Actions (CODA) model. Instead of a fixed probability that the other agents will decide in the favor of the best choice, each agent considers that other agents might be one one of two types: trustworthy or useless. Trustworthy agents are considered more likely to be right than wrong, while the opposite holds for useless ones. Together with the opinion about the discussed issue, each agent also updates that probability for each one of the other agents it interacts withe probability each one it interacts with is of one type or the other. The dynamics of opinions and the evolution of the trust between the agents are studied. Clear evidences of the existence of two phases, one where strong polarization is observed and the other where a clear division is permanent and reinforced are observed. The transition seems signs of being a first-order transition, with a location dependent on both the parameters of the model and the initial conditions. This happens despite the fact that the trust network evolves much slower than the opinion on the central issue.