Hei Erlend. Nice read. It's probably true. I use LLMs at work, because that's the job now. I also use them on some side-projects, and in others more sparingly, simply due to a decade of knowledge standing in the way of what is currently fairly good but not always there LLM generation. However, I am under no illusions that speaking about todays LLMs is not going to match the capability of next years LLMs. Currently, for projects where I am an expert, reviews and benchmarking, and debugging are all helping greatly. I am perfectly fine nitpicking details and occasionally getting angry at "obvious pessimizations". The problems I see are sometimes just a misunderstanding. In some places correctness is preferred, in others speed is everything. So, it's essentially me correcting my own lack of upfront specificity. For other things, like CI, I'm quite happy to just let the tool do the job and leave it at that. I will inspect the artifacts and do manual testing afterwards. I think a tool is still what it is, and I don't quite understand where the anthropomorphization is coming from. Or, the anger from the other side where only hand-written code is good, when it clearly isn't (and never was). I've written enough bugs to know having an full review in less than 10 minutes is game changing. Or even that we sometimes have to leave aging languages behind, and embrace new ones, or we will just become the programmings equivalent of "science advances one funeral at a time".