This paper argues that the CHREST architecture of cognition can shed important light on developing artificial general intelligence. The key theme is that "cognition is perception." The description of the main components and mechanisms of the architecture is followed by a discussion of several domains where CHREST has already been successfully applied, such as the psychology of expert behaviour, the acquisition of language by children, and the learning of multiple representations in physics. The characteristics of CHREST that enable it to account for empirical data include: self-organisation, an emphasis on cognitive limitations, the presence of a perception-learning cycle, and the use of naturalistic data as input for learning. We argue that some of these characteristics can help shed light on the hard questions facing theorists developing artificial general intelligence, such as intuition, the acquisition and use of concepts and the role of embodiment.