Systematic genome sequencing is opening up new research frontiers across the life sciences. But how do you approach a hyperdiverse group like insects? We used to believe there were about 5.5 million insect species on Earth, only 20% of which had been described. However, recent inventories using DNA metabarcoding suggest that the total diversity is in the tens of millions, and that our knowledge of the insect tree of life is even more fragmented and skewed than previously thought. Current genome sequencing efforts tend to amplify rather than mitigate these biases, seriously compromising the goal of obtaining a representative picture of insect evolution. We will address the missing pieces in the insect genome tree using the unique global insect material (35,000 samples, 30 million insects, 1 million species) collected by the Insect Biome Atlas and LIFEPLAN projects. Because the species content of each sample is indexed through metabarcoding, we can bypass traditional taxonomy and systematically build a representative insect tree. We will develop innovative approaches to all steps in the workflow, from visual detection and identification of target specimens with machine learning to probabilistic machine learning for tree inference. We will pursue breakthroughs in five research directions: (1) probabilistic analysis of combined genomic and metabarcoding data at unprecedented scales; (2) machine learning of the links between genomes and phenomes across the full diversity of insects; (3) first analyses of the true diversification of insects across geological time; (4) discovery of new evolutionary processes and genomic signatures in previously unexplored parts of the insect tree; and (5) novel insights into the diversity and evolution of insect symbionts. We will also build a new foundation for assessing the state of global insect biomes, from automated identification of species to improved accuracy of biodiversity impact reporting using environmental DNA.