I’d probably ask for a much better metagenome assembler and binner. there’s enough groundwork and data out there with exponentially more created each year. If an AGI could compress the 20 or so PhD student-years needed to really leverage that into something outstanding into a few hours that would be really nice.