Agent-based modeling (ABM) is a simulation technique which has been increasingly integrated into the economic discipline in order to understand complex systems. However, most of everyday research activities rely on the researchers' consensus concerning practical choices about modeling strategies, computational boundaries under scrutiny and the extent of empirical validation. Particularly lacking are reflections on the semantic construction of conceptual models. The paper reviews existing theoretical frameworks leading to understanding ABM as a technique, where the cognitive processing instantiated by the instrument is distributed across different modeling layers, including conceptual, algorithmic and computational ones, which can be interpreted as an interlinked set of analogies. Then, it introduces a framework for assessing ABM conceptual adequacy and tests it on two families of models in the field of economics of innovation, revealing several modeling constraints.