This papers contributes to the stream of research on rule based behavior, and rationality. A bounded rational agent can deal just with a reduced number of variables, neglecting part of the overall complexity. This is usually taken as just a limitation: agents cannot deal with all relevant information and use biased decisional shortcuts. The stream of research on Ecological rationality, yet, evidences the possible advantage of using a limited amount of information. The present paper takes a similar, but not identical, point of view. I propose an idea based on some contributions on the ecology of the mind by Gregory Bateson. Learning requires to recognize a series of situations as identical and then to observe the effect of given variables in specific fixed contexts. Two situations can be considered identical only limiting considering part of the overall information and taking as unchanged a series of factors. This process determines an individual representation which have just to be coherent with the world. Only in abstract world contexts are objective situations. In the real world, they are just hypothesis to be continuously tested. This vision of bounds and learning has many implications for the debate on rationality and rule following.