The evolution of antibiotic resistant bacteria threatens to become the leading cause of worldwide mortality. This crisis has renewed interest in the practice of phage therapy. Yet, bacterial capacity to evolve resistance is likely to debilitate this therapy as well. To combat the evolution of phage resistance and improve treatment outcomes, many have suggested leveraging phages ability to counter resistance by evolving phages on target hosts before using them in therapy (phage training). We found that during in vitro experiments, a phage trained for 28 days suppressed bacteria ~1000-fold for 3-8 times longer than its untrained ancestor. This extension was due to a delay in the evolution of resistance. Several factors contributed to this prolonged suppression. Mutations that confer resistance to trained phages are ~100x less common and, while the target bacterium can evolve complete resistance to the untrained phage in a single step, multiple mutations are required to evolve complete resistance to trained phages. Mutations that confer resistance to trained phages are more costly than mutations for untrained phage resistance. And when resistance does evolve, trained phages are better able to suppress these forms of resistance. One way the trained phage improved was through recombination with a gene in a defunct prophage in the host genome, which doubled phage fitness. This direct transfer of information encoded by the host but originating from a relict phage provides a previously unconsidered mode of training phage. Overall, we provide a case study for successful phage training and uncover mechanisms underlying its efficacy.