Annotating the genes in a newly sequenced genome

The first bacterial genomes, in the middle of the 1990s, were annotated gene by gene by teams reading alignments and arguing about start codons. Sequencing then outran reading by several orders of magnitude, and structural annotation, which is to say deciding where each gene begins and ends, became a pipeline job that nobody proposes to do by hand again. What was ceded is that structural pass, completely for bacteria and very nearly for large genomes. Not ceded: the reference sets. The human gene set is still built by merging automatic models with manual curation, because curators see transcripts the pipeline misses, and because a functional claim about a gene is an assertion somebody has to stand behind.