Standard supervised classification trains models to strictly mimic the exact labels of a perfect oracle. Since this typically happens in a single forward pass, models are locked into a fixed compute budget whether an input is simple or highly complex. Moreover, this rigid training objective forces the model to express absolute certainty on its training data, which often carries over into evaluation, leading to overconfident predictions. To address these limitations, we propose Reinforced Iterative Classification (RIC), which replaces the imitative objective with Reinforcement Learning (RL). RIC deploys a recurrent agent that iteratively refines a predictive distribution over classes, receiving reward for stepwise improvements in prediction quality. Using image classification as a testbed, we show that RIC maintains the accuracy of supervised baselines while achieving better calibration. The learned policy naturally manages its own computation by allocating more effort on resolvable inputs while halting when further improvement appears unlikely.