Agent based design systems could provide useful decision help for architects working on spatial planning tasks that involve large number of actors or deal with complex urban situations. These systems are especially helpful in bridging the gap between concrete design proposals and high-level design abstractions such as frequency and flow diagrams. Every attempt to use computational design agents in the planning process will automatically raise many fundamental issues about spatial perception and representation of the environment. The paper discusses these issues in the light of some recent agent based simulations. Two case studies are presented in order to demonstrate different uses of computational agents in urban design. The first study shows how a simple agent-based design system placed in an urban context becomes a creative production tool. The second one reveals analytical capabilities of an agent system in urban environments.