

I use AI at work. The use of AI requires knowledge and projects structured around the use of AI. If you do it properly, AI code is much better (100% clean code architecture adherence, tdd with >90% coverage, deploy gates on quality and security, easy dependency upgrade). Since we started using AI properly our codebase quality skyrocketed. Less bugs, more features, less insecure code. You do not know how to use AI properly. That is fine, most don’t, we are discovering how to use a new tool that change every week now. That does not mean the problem is the tool.






Fine, but most people talk about AI, and AI datacenter, and AI coding to refer to LLM, so it is what I use, well knowing we are talking about LLM. And while I grant you that the entire software development world is not ready for LLM, having spent decades optimizing for people writing software, there is a lot that we learned in the last year of consistent use applied to software development.
LLM exhibiting true intelligence or the appearance of intelligence is inconsequential if the final result mimics the reasoning capabilities of humans working on software.