Over the last decade, artificial neural networks enabled substantial progress in artificial intelligence. These networks were inspired by an understanding of biological neural networks which is today largely outdated. In this talk, I will present several ideas from neuroscience that I believe will be important for designing agents that can achieve complex goals. These ideas include the modularity of brain circuits, cognitive maps, and consciousness. I will also speak about my current project whose aim is to demonstrate a simple proof of concept of a planning agent which will be based on these ideas.