In the article we discuss some questions connected with cognitive structures and behavior algorithms of an artificial social agent. We suggest a cognitive Plan-Image architecture that is based on the concepts of TOTE model by Miller, Galanter and Pribram. In order to evaluate efficiency of the architecture we use a classical search task on a two-dimensional grid. Each agent has a map that reflects his current information about the location of objects (Image) and alternative search algorithms (Plans). Objects in the grid may be arranged randomly or according to a certain law even or centered pattern. The experiment aims to assess how effectiveness of the search changes depending on the amount of available information and agents search algorithms.