The architecture of FCG and Babel has a double-layered design. The first layer is called the routine layer and handles habitual processing. A second layer, called the meta-layer, monitors and sometimes steers routine processing through diagnostics and repairs, which try to detect and solve problems that may occur in the routine layer. Repairs have the power to modify an agent’s inventory of concepts, linguistic constructions, beliefs, and so on. They can also go back a few steps, for instance choosing a different communicative goal or parsing an utterance again, in order to test whether the repair adequately solves the detected problem. Problems can be detected at each level, at each step and at any time; and different repairs can be triggered in succession of each other. This constant interaction between routine- and meta-layer processing ensures robustness and open-endedness for coping with noise or variation in perception, differences in embodiment, novelty, and other problems that inevitably occur in linguistic interactions.