All large-scale structure cosmologists are faced with the question: how do we robustly extract cosmological information, such as on dark energy, gravity, and inflation, from tracers such as galaxies whose astrophysics is extremely complex and incompletely understood? Based on advances in our theoretical understanding of large-scale structure, we now know how to absorb these complexities into free "nuisance" parameters on large scales. This opens up the possibility for physically robust cosmology inference from galaxy clustering. However, to really make use of the power of this approach, we have to go beyond current analyses based on the power spectrum. I will describe a novel approach that attempts to extract the information in the entire galaxy density field, rather than compressing it into summary statistics, while robustly marginalizing over the complexities of galaxy formation.