At this point it's nice hacking, but not very productive yet. It needs to become more proactive and autonomous, while being conscious of token limits, or hopefully running on decent local models soon. But I think we'll have actually useful and easy to set up agents later this year.