What’s been on my mind is how people are handling the parts where AI really isn’t that reliable yet — a GDPR-compliant privacy policy, Terms of Service, the EU AI Act, or processing customer payments.
Makes me wonder if owners of these vibe-coded apps actually think about this before launching and bring someone in to help, or just take whatever AI spits out and push it straight to production.
Most likely this….seems even intelligent people are letting AI run the show and believe it actually is artificial intelligence and not just the median output from all training inputs.
I imagine most never get enough users to uncover the broken bits anyway.
I recommend you take a quick look around vibecoding heavy subreddits, and vibecoding tool subreddits and you’ll find an insane amount of pissed off users.
The messaging is: build anything. The reality is: you need $3k in tokens and good luck writing the correct prompts or it’ll go up to $5k. And if you are lucky you’ll get something usable.
Vibecoding turns apps into blackboxes, and the more you prompt the more cognitive debt the build accumulates.
I honestly don’t know how people spend so much on tokens. I use Codex to build a lot and I never run out. Until recently I was on their $20/month sub and now upgraded to $100/month. There are limits, but I’ve never run out.
But I do see token usage increase rapidly when I use V0. And I guess that is because you are paying a margin and the usage is truly metered. So the lesson getting closer to the source (get better and proper skills) and you won’t need to blow so much of your budget.
I guess some people are also relying on the AI to fill in the gaps in their knowledge rather than directing the AI properly and surgically.
This is probably why the subreddits are full of people complaining.
I’ve seen a ton of upset people, as was mentioned.
I think it’s important to remember, though, that most of the people building an app with AI these days have no real idea of coding or anything close to that.
You use Codex, and other experienced people use similar platforms, but to the average business owner or someone with an idea for something, they’re lost with some of these platforms.
Bubble is one of the few nocode tools (if not the only) tool that allow you to actually SEE what’s going on.
It’s for me a 100% valid reason to stay here.
Elsewhere, what you get is at most a super sexy prototype. As soon as you get users and real life usecases, it leaks from everywhere.
And what about token usage: vibe-coding from the sofa through infinite long threads costs lots of token. Each single message makes the chat heavier and heavier, creates hallucinations and costs money exponentially.
Very reason why Bubble is the right tool for the new age of “code on demand”.
Bubble has been the best starting point for non coders it still is. As much as people hate on Bubble for working so much on the AI builder, in the long term Bubble’s value still comes from the editor and it’s visual language.
Buildprint is also proof that experienced builders can leverage AI to maintain and optimize Bubble apps.
Bubble isn’t perfect but IMO it’s heading in the right direction. It does however need to continue to build bridges between what it’s built and lower level code. AI is already a great guide between the 2 world’s.
Good examples of Bubble heading in the right direction currently in regards to bridges:
Bubble native JSON parsing
True if else functions in expressions
Code languages are abstractions to make it easier for developers to understand how software interacts with machine, Bubble is one of the best abstractions to visualize it.
