In contrast to deep-learning agents, biological neural networks are grown through a self-organized developmental process. In this work, we propose a new hypernetwork approach to grow artificial neural networks based on neural cellular automata (NCA). Inspired by self-organising systems and information-theoretic approaches to developmental biology, we show that NCA can be used to grow neural networks capable of solving standard reinforcement learning tasks. Finally, we explore how the same approach can be used to build developmental networks capable of metamorphosing their weights to solve variations of the initial RL task.