A compact multiagent optimization system (MAOSC) based on autonomy oriented computing (AOC) is presented. Performed by a society of autonomous entities in iterative cycles, an optimization algorithm can simply be described by a macro generate-and-test behavior, which deploys a few elemental generating behaviors under conditioned reflex behaviors supported by a testing operation library. MAOSC provides a simple framework for not only realizing and comparing algorithms, but also deploying evolvable algorithms. The experimental results of MAOSC cases on benchmark functions are compared with those of other algorithms,which show its efficiency.