I summarize and analyse key problems with observations of the UV bump in active galactic nuclei (AGN), and especially the accretion disc interpretation - the temperature problem, the ionization problem, the time-scale problem and the co-ordination problem - and suggest that all these problems can be solved if, in addition to the accretion disc, there is a population of cold, thick clouds at approximately 30R S which reprocess the intrinsic continuum. Exploring cloud parameter space, I find that clouds with density n∼ 1012 cm−3 and column N H > 4 × 1024 cm−2 reflect most of the intrinsic continuum, but convert a substantial fraction of the EUV luminosity into lines, dominated by Lyβ and He ii Lyα. When velocity-blurred, this makes a false continuum peak at ∼1100 Å which fits the observed spectral energy distribution (SED) well, but turns back up in the FUV to make a hard EUV SED, as required by ionization models. I argue that the observed UV variability is dominated by this component of fixed shape, possibly due to changes of covering factor. The amount of mass required is small, so it is not necessary to disrupt the disc, but only to make an unstable and inhomogeneous atmosphere. The proposed clouds may be related to those suggested by several X-ray phenomena (X-ray reflection components, high-velocity outflows, Compton thick partial covering) but are not the same, leading to a picture with a wide range of inhomogeneous structures at different radii.