Agent swarms, alignment, & recursive self-improvement