Reinforcement learning is a promising artificial intelligence area which relies on computational simulations where an agent finds a policy that maximizes a particular reward by mapping states to actions in a non-deterministic environment, with applications in optimization problems. Here, we apply Q-learning, a popular model-free reinforcement learning algorithm to a representative ecological problem: the sustainable harvest of a population which follows a logistic growth rate by modeling it as a valid python gymnasium environment, contributing to the generation of new environments which represent real-world situations and may lead to decision making. We implemented three different scenarios in which the agents determined an optimal policy guided by different reward signals: a greedy agent, which harvests as much as possible; a conscious agent, which seeks to harvest as much as possible while maintaining high population numbers; and a biocentric agent, which harvests only what is needed to avoid the population to exceed the carrying capacity of the environment. By initializing the population size to 100, with a growth rate of 1 and a carrying capacity of 200, the biocentric agent had the largest accumulated harvest, suggesting that a sustainable strategy is more profitable in the long term. Reinforcement learning is a very promising approach for decision making in many ecological real-world situations, where population management is a very straightforward application.