In the mouse blastocyst, fibroblast growth factor (FGF) signalling regulates the differentiation of inner cell mass (ICM) cells into epiblast or primitive endoderm (PrE) cells. FGF was believed to drive spatial patterning, but Archishman Raju and Eric Siggia now re-examine published data and show that cell movements in the ICM mean FGF exposure can be considered spatially uniform, but time dependent. Thus the authors propose a geometric model of ICM differentiation, in which the Waddington landscape analogy of cell differentiation is extended from a single cell to the whole ICM. They contrast two possible models. The ‘dual cusp’ landscape proposes that a direct bifurcation decision is made by ICM cells to differentiate into epiblast or PrE, which cannot switch fates except via the ICM. In the second ‘heteroclinic flip’ model, differentiated cells can directly transition between epiblast or PrE in response to the timing of FGF. By testing the model with experimental data from FGF perturbation experiments, the researchers argue that the ‘heteroclinic flip’ model best explains the temporal effect of FGF signalling on differentiation and provides a model by which morphogen feedback can direct differentiation at a population level by controlling relative proportions of cells.