We define an agent-oriented abstraction formalism devoted to generalized theories of abstraction that have been proposed in Artificial Intelligence. The model we propose extends the abstraction capabilities of the existing Agent-Oriented Programming paradigm. This short note reviews first the existing attempts to define abstraction in AI and in agent systems. Then, our model is introduced in terms of six definitions covering the concepts of agents, annotated knowledge, utility and society of agents. The presentation is purely cognitive and intuitive.