Agent-to-agent pair programming