10x easily. 100x if you're a wizard.

So, hypothetically speaking, suppose I had been working on something for the past 8 years. It could’ve been finished in less than a year, with today’s technology?

Yep. But this also means scaling the bugs, the slop, the security holes, and the bad UX with it. Can’t have it both ways. The AI was trained by public code and human thinking from the internet. Have you been on the internet?

No, it could be started faster. AI falls off with the project scale, at least for now. Errors accumulate, and global understanding of the code base worsens. Still should work well for minor junior-level tasks or code conversion between languages, if the source code is good and working then you'd probably only need to fix a few bugs after conversion which is faster than learning both languages from scratch. At some point explaining to the model what you want and kicking it to fix errors might become less productive than doing it yourself. And the worst mistake is fully trusting the AI output without understanding and checking it yourself. That's where the huge problems originate, because these models always 100% sure they're correct until you corner them, and even then they might insist you're wrong. But hey, this vibe-stuff only makes human programmers more valuable as we're hired to fix the bugs and security holes! 😂