Cities are among the largest and most complex artifacts created by humans. Yet the advancements in computing capabilities combined with ubiquitous data streams allow for complex socio-technical systems of cities to be abstracted and modeled. This paper discusses the technology of Smart City Digital Twin as a policy tool. Following the ideas of flat ontology, the paper argues that intelligent machines exhibit their own agency, which has to be investigated through the behavioural lens. As making policy decisions based on counterfactual simulations is becoming more widespread, it is crucial not only to simulate how certain policy interventions will affect the life of a city, but also to investigate how such models and simulations are designed and behave. Adding a social layer in the form of behavioural data of population will allow Smart City Digital Twins to be used for a wider spectrum of policy modelling purposes. Such behavioural data can be generated through task-based approach, where individuals will be asked to conduct certain activities in order to generate synthetic data for situations which require data that does not yet exist. This will not only allow to avoid certain privacy-related concerns, but also can be used as a tool for labor provision.