Current programming AI is a multiplier of what a human can already do in programming combined with what he knows about how to use that AI [A quarter of startups in YC's current cohort have codebases that are almost entirely AI-generated | TechCrunch](https://techcrunch.com/2025/03/06/a-quarter-of-startups-in-ycs-current-cohort-have-codebases-that-are-almost-entirely-ai-generated/) [https://www.youtube.com/watch?v=riyh_CIshTs](https://www.youtube.com/watch?v=riyh_CIshTs) "A quarter of startups in YC’s current cohort have codebases that are almost entirely AI-generated the most I've seen so far when it comes to primarily AI-generated code being deployed into production is in Silicon Valley in Y combinator startups, where a quarter of them already have 95% of their codebase AI generated code, but even though a lot of them are trying cool innovations, I think there will be a lot of problems at the same time i think a lot of those people are using it too heavily too prematurely and there's going to be a lot of security bugs and hard to maintain structures that are going to bite them in the butt soon but the other question is whether or not will AI improve even faster this year than it did last year so that it can clean up those bugs and technical debt, otherwise I think a lot of these people are in danger of an infrastructure implosion a lot of things are getting solved and the capabilities of those systems are accelerating rapidly, and a lot of people are doing a lot more with it than they were able to a year/few years ago, but some fundamental limitations of the existing architectures still exist where I think there's not enough innovation like I use it every day at work and in my spare time too for everything and it's absolutely awesome and the progress is insane (today the new Gemini 2.5 Pro model gave me more mindblowns), but some people just don't know what a for loop is and then are surprised that they're totally screwed when the AI can't fix something yet for now and they don't know how the code works and aren't able to learn it fast enough when they're so far without knowing the basics of programming to me at this point in time in most domains it's primarily a pure amplifier in those different problem domains, and depending on what skills and knowledge one has about those AI systems and how much one knows about the problem domain one is using it for, one can dial the degree of delegation to AI of those different parts of the problem that one is solving there are domains where it's already at the level of humans or superhuman (like playing some games like chess/go/dota/poker, classification in some contexts, recommending content on social networks, etc.), but it's not really in software engineering yet :D" Someone should make a meta tool that calls all vibecoding programming scaffolding systems and then does agentic megadiscussion about the result