Literally none of the predictions you listed in your post came true. We don't have reliable agents, AI isn't writing most code, hallucinations still happen all the time, improvement has been basically non-existent. Despite the constant claims of these experts and AI boosters, AI is still not a tool one can use to get meaningful work done.
This is called denial.
You're in denial. The great problem with AI coding is now that it does too much, not too little. See the endless stream of vulnerabilities or the constant barrage of vibe coded apps, that get ever more complex.