During the last decade, both evolutionary computation and multi-agent systems have been used for solving decision and optimization problems. This paper proposes a new evolutionary agent system by incorporating evolu-tionary process into agent concepts for solving mathematical programming models. Each of the agents represents a candidate solution of the problem, and able to sense and act on the society. The fitness of the agent improves through co-evolutionary adaptation of society with the individual learning of the agents. The performance of the proposed algorithm is tested on five new benchmark problems along with existing 13 well-known problems, and the experimental results show convincing performance.