This is my observation based on working with AI. AI seems to mature just like people mature. It has progressed from an intern into a staff or senior level engineer. I still see it making bad decisions on security (mostly errors of omission) and the quality is still greatly dependent on the the instructions. I do think this continues to get more advanced over time. What I am looking forward to understanding is what does a code base look like that's been 99% developed by agents over a long period of time. These early agents may be technical debt engines at scale.
