In recent years, the CMS experiment has achieved significant progress in the reconstruction and identification of hadronically decaying tau leptons, largely driven by advances in machine learning techniques. The current CMS standard employs the Hadron Plus Strips algorithm for hadronic tau reconstruction and the latest DeepTau algorithm, based on Convolutional Neural Networks, for the identification step. This proceeding summarizes these methods and presents a comparison with a novel approach that relies entirely on jet objects. The new method extends existing jet flavor tagging techniques, based on Graph Neural Networks, to also include the identification of hadronically decaying tau leptons. The unified taggers presented here, PNet and UPart, demonstrate several advantages over the traditional approach, while also exhibiting certain limitations. By combining the strengths of old and new algorithms, it is possible to achieve the best identification performance of hadronically decaying tau leptons to date.