i have the same thought if you get good enough intuition on building long-horizon RL tasks on particular domains
the field is maybe fairly new for high schoolers and undergrads but you will be champ once you start looking into agent traces and being creative.