<p>Over the past decade, the search for new thermoelectric materials has shifted profoundly, from reliance on experimental serendipity to computationally guided discovery. This transformation is particularly challenging due to the complexity of the terms governing thermoelectric performance, including dopability, electronic structure, scattering mechanisms, and thermal conductivity. In this article, we frame thermoelectric discovery as a two-stage design challenge: identifying compounds with high thermoelectric potential and then optimizing their performance through dopants, defects, and alloying. The first stage benefits from automated screening pipelines based on density functional theory, while the second relies on finer-grained models that capture local structural motifs, complex scattering mechanisms, and alloy thermodynamics. Advances in machine learning and artificial intelligence have revolutionized every stage of this design process, providing inexpensive surrogates, uncertainty quantification, and access to mechanistic insights.</p> Graphical abstract <p></p>