Synthesis of subagents: exercise — LessWrong