I will show how the non-Gaussianity induced by the non-linear growth of structure needs to be accounted for for the next-generation of galaxy surveys. I will present new non-Gaussian covariance terms and their numerical implementation for the galaxy angular power spectrum, showing that it is necessary to go beyond the previous state of the art already for a Euclid baseline cosmological analysis. This non-Gaussianity will prove critical to avoid unrealistically optimistic forecasts on Dark Energy constraints, but I will also show that it has positive consequences. Pushing the analysis to smaller scales, I will challenge the community's unconscious lore, by showing that information disappointingly plateaus at the beginning of the non-linear regime, but strikingly rises again in the deeper highly non-linear regime. I will carefully explain why this happens, due to the impact of non-linearity both on the observables and on covariances. I will conclude on how this motivates joint analysis of cosmology and astrophysics.