This is one of the few agent abstractions I've seen that actually seems intuitive. Props to the OpenAI team, seems like it'll kill a lot of bad startups.