Why do you assume any point in time criticism of AI code generation is reliable to project even a couple of months into the future?
Current models are in fact much better than the top models from a year ago. I have no idea how anyone could claim otherwise without a huge amount of motivated reasoning.