Well, it depends. They are better programmers than me and most programmers in the world when it comes to heavy-lifting. They produce code practically instantly. I don't need to know the details (unless I want to), I need to guide it on the path. Most other programmers in its stead would take longer to do the job, despite having a better sense of following the direction and long term development (but AIs are growing on that side very quickly). There is one limitation: I don't give it access to my computer. Had it had access to files and the ability to troubleshoot the results of its code and test it against real data as it progresses, it would probably do better, with less interventions from my side. I also don't use the latest model because it's not free, and I have my reasons not to pay for it (mainly, in my opinion, they become addictive, and the free models have limits that are actually good for us to take long breaks from AI and switch to something else).