Over the past five and a half years, SARS-CoV-2 has demonstrated in real time many concepts and principles of evolutionary biology. Soon after it was disseminated globally, the virus underwent adaptive radiation resulting in the generation of multiple dominant variants. Later variants drove earlier ones to extinction in a series of selective sweeps. The nature of adaptation was shifting molecular specialization, with the spike protein losing binding affinity toward bat cells to gain affinity toward human cells, losing replicative fitness in lung cells to gain fitness in nasal cells. Evolution of the spike mutations was constrained between two goals, enhancing receptor binding and evading neutralizing antibodies. Because there are limited ways to achieve the two goals, multiple variants converged on the same spike mutations, with higher-impact mutations fixed before lower-impact mutations, giving a new meaning to diminishing-returns epistasis. Later genetic changes became repetitive and cyclical. The Delta variant represented an evolutionary dead end. Evolution of the virus also demonstrated punctuated equilibrium, with saltatory changes producing highly mutated variants which subsequently experienced gradual structural and functional drifts. While structural proteins experienced strong positive and purifying selections, nonstructural and accessory proteins accumulated neutral and deleterious mutations, most of which remain unfixed. Selection of adaptive missense mutations resulted in deoptimization of codon usage. These phenomena point to Muller’s ratchet in action. The higher codon usage score in the initial Omicron variant was probably due to long-term preservation of the virus in an immunocompromised host where low immune pressure prevented genetic degradation.