What we can learn about human-agent teamwork from practice