Vibe Coding: When AI Writes, but Doesn’t Build
At first, it was chaos. Then it started working. You wrote two lines, and it built a house of cards. Then suddenly, chaos again.
With multi-agent setups and refined model versions, it looked stable. Even complex projects seemed to run fine, until the first LLM update and everything broke again. The internet is now filling up with apps and websites that seem to work until you touch them. One version change, one unexpected command, one small deviation, and everything stops working.
It’s like building on sand: AI generates code that only holds until someone actually uses it. Once you deploy it in the real world, in production, with real users (maybe a lot of real users), it collapses. Because there’s no real architecture, no robust logic, just the illusion of intelligence.
Here’s what happened: anyone with an idea could turn it into an app or a site in days. A market drug. Some experts saw the limits immediately. Others, beginner programmers new to code, were mesmerized. Something that had always been impossible now seemed to come alive. Apparently.
Now even Chamath Palihapitiya and Andrej Karpathy, the one who coined “Vibe Coding,” are admitting its limits. But Vibe Coding remains a tool, and like any tool, it depends on how you use it. If you let it do everything, if you don’t know what happens behind the scenes, if you can’t code or design architectures, it will fail. But if you can code, if you know the logic, and you use Vibe Coding as an accelerator, you can make those houses of cards stronger.
Vibe Coding isn’t good or bad. It’s just a tool. What matters is how well you understand it, how much you test it, how much control you keep. This video is meant to show it for what it really is: not magic, but an unstable mode we can learn to understand.
Not by chance this series is called Artificial Decisions. Because in the end, the ones who decide what to build should still be us.
#ArtificialDecisions #MCC #AI