I use claude code daily, yes it gets things wrong but the stuff it's good at it's ridiculously good at. I have to go in and check, but I still don't write any code manually, I'll just guide it in the right direction. Whenever it gets something wrong it will just fix it's own work during dev testing or when the testers get around to finding them. Bug fixing is ridiculously quick and easy. I feel like I'm just a manager really at this point. The problem is it's so fast all of our deadlines are based on using claude. Meaning if I said fuck it I'm writing it myself, I'd fall behind fast and would miss all sprint targets. 

Eventually it'll need less and less people to run until 5-10 years from now dev departments will just be a handful of people (as someone needs to be accountable for when things go wrong). Token cost is a massive issue which might save us, so hopefully it doesn't really get much cheaper and we're still needed. Cause at the moment it's the only thing stopping my team from being 1 person with 6 claude sessions running simultaneously. 

Unfortunately I feel more pessimistic than not about AI. There isn't enough being done by governments worldwide to ensure that human workers who are laid off because of AI are able to make ends meet. I also think AI will continue to worsen problems that have existed because of the internet and social media. People will dig deeper into spaces that align with their own views instead of having deep conversations face-to-face with people they may not see eye to eye with. As a gamer, the massive increases in RAM pricing due to AI data centers are another pain in the ass.