Vibe coding’s code is fine—the bill is the problem

Arpit Bhayani said on X: No human can write better code than an LLM. Accept the fact and stop pretending.

Suryansh Tiwari quote-tweeted and agreed, but added one point: That's true, because writing code was never the hardest part. This week, apps built with vibe coding are falling apart everywhere.

Vibe coding means writing code by intuition without digging into the underlying logic. This development style has become popular recently, and the costs are starting to show:

• A business website was hacked six days after launch, with the admin password exposed directly in the page source.
• Someone joked that they used vibe coding to build an email tool and ended up with a $4.4 million AWS bill.
• Developers are starting to warn each other: an app rushed out this way may not close its first deal, but might receive a $100,000 legal letter first.
• Another developer built a game with AI but had no idea how to deploy it online.

In these cases, the code itself runs.

The problem is everything around the code: where secrets are stored, who has permission to log in, what to do when traffic suddenly spikes, who pays for the money burned by an infinite loop, and which machine the app is actually running on.

Engineers spend years learning exactly these things. Vibe coding skips these steps, then runs into them one by one after launch.

Gaurav Sen calls this "system design over syntax." Suryansh says it's actually the boring 80% of the work.

Aniketh posted this image in a reply, making the same point:

All implementations are done by LLMs, then cross-checked by LLMs from different providers. But the guards remain: fuzzing, verification, and differential testing—none can be skipped. LLMs handle the writing; humans handle knowing where the boundaries are.

Suryansh is a growth marketer, not an engineer. When he builds with AI, he chooses an environment where all the rules are already set up.

That's what Floot is doing. Secrets aren't embedded in the code but managed separately; login and databases are preconfigured; every app has a daily hosting budget cap, so even if a process gets stuck in an infinite loop, the bill won't roll up to $4 million.

Code has become cheap. The hard part is knowing what to build and what must be true before launch. The latter still requires repeated practice.