Current phylogenomic methods are still a long way from implementing a realistic genome evolution model. An ideal approach would require a general joint analysis of genomic sequences, while including coding sequence annotation, protein evolution or gene transfer, among other mechanisms , to infer the complete evolutionary history of the studied genomes. Such an approach is computationally intractable and currently approximated by phylogenomic pipelines that implement a series of independent steps ranging from gene annotation to species tree inference or positive selection detection. Here we review the virtues and limits of current phylogenomic methods compared to what could be expected from an ideal method. We present five case studies to illustrate various issues and limits in current phylogenomic practices, while assessing their relative importance. We argue that data error is pervasive in modern datasets and models are still too simplistic compared to the complexity of biological and evolutionary processes. Importantly , joint analyses should be a research focus as the many steps of phylogenomic pipelines are not mutually independent. It is essential to recognize the hidden assumptions of the many types of analysis available to our community so as to circumvent model misspecifications and critically evaluate the relevance of their results. In conclusion, the quality of datasets should be enhanced via numerous, rigorous checkpoints, while also boosting the capability of models to handle biological complexity by the development of better models, particularly through joint analyses.