Semantic 3D city models are one of the cornerstones of the so-called "smart city" applications, yet they are very difficult to manipulate/edit/update. While they are relatively easy to generate, their use and maintenance is limited by the available software and the cumbersome data model. Have you ever tried to write a CityGML file? And to parse one? We did, and we didn't like it. Therefore we created a developer-friendly JSON implementation of the CityGML data model. This talk will introduce CityJSON and its processing software cjio, which can be used as a CLI to chain operations, or its API can be used to generate features for machine learning.