Here are some examples of the use and study of agent-based, self-organizing systems in architecture:
The images above are from the project Swarm Urbanism (2008) from Kokkugia, a speculative proposal which
(...) posits an urban design methodology based on the emergent capacities of Swarm Intelligence in rethinking of the current redevelopment of the Melbourne Docklands. Swarm systems involve the local interaction of autonomous agents, which give rise to emergent behavior and the self-organisation of structures. An application of swarm logic to urbanism enables a shift from notions of the master-plan to that of master-algorithm as an urban design tool. This shift changes the conception of urban design from a sequential set of decisions at reducing scales, to a simultaneous process in which a set of micro or local decisions interact generate a complex urban system. Rather than designing an urban plan that meets a set of criteria, urban imperatives are programmed into a set of agents which are able to self-organise. Consequently this conception of urbanism generates systems that are flexible to respond to the constantly changing political, economic and social pressures of urban development.
Also from Kokkugia, the above project (Emergent Fields, 2003), in which
(...) Agent-based simulation techniques are used to generate programmatic relationships and an architectonic response to this field of program. Architectural elements such as a façade, plaza, or construction grid are assigned rules or behaviours, which govern the way in which they interact with this field in the form making process. This develops an emergent relationship between program and peculiarities of architectural form, enabling the design process, and resultant architecture to exhibit particular behavioural qualities.
Also in the research developed by Space Syntax together with the Space Group at Barttlet, we can find the use of agent-based systems to explore pedestrian behaviour within the field of urbanism, flight-routes, programmatic organization, etc.
Space syntax has found that, despite many proposed higher-level cognitive models, there appears to be a fundamental process that informs human and social usage of an environment. In this paper we describe an exosomatic visual architecture, based on space syntax visibility graphs, giving many agents simultaneous access to the same pre-processed information about the configuration of a space layout. Results of experiments in a simulated retail environment show that a surprisingly simple 'random next step' based rule outperforms a more complex 'destination based' rule in reproducing observed human movement behaviour. We conclude that the effects of spatial configuration on movement patterns that space syntax studies have found are consistent with a model of individual decision behaviour based on the spatial affordances offered by the morphology of the local visual field.