Imposing key anatomical features, such as the number of organs, their shapes, sizes, and relative positions, is crucial for building a robust multi-organ segmentation model. Current attempts to incorporate anatomical features include broadening effective receptive fields (ERF) size with resource- and data-intensive modules such as self-attention or introducing organ-specific topology regularizers, which may not scale to multi-organ segmentation problems where inter-organ relation also plays a huge role. We introduce a new approach to impose anatomical constraints on any existing encoder-decoder segmentation model by conditioning model prediction with learnable anatomy prior. More specifically, given an abdominal scan, a part of the encoder spatially warps a learnable prior to align with the given input scan using thin plate spline (TPS) grid interpolation. The warped prior is then integrated during the decoding phase to guide the model for more anatomy-informed predictions. Code is available at https://anonymous.4open.science/r/AIC-UNet-7048https://anonymous.4open.science/r/AIC-UNet-7048.