Don’t forget the copy/paste doesn’t work well…sometimes tons of issues. Even get editor freeze if going from one tab to another tab sometimes.
If they do sandbox, must be on same editor page. And even then, there will be the insert issues potentially.
Yep, separate problem which I’ve mentioned before. They’re trying to build on top of a shaky foundation.
If there are issues with moving/splicing/combining things programmatically, how do they expect a non-deterministic system to fare better? It’s only as strong as the programmatic guardrails.
@randomanon honestly I just did a side by side Lovable vs Bubble AI (with the new Sonnet 4.6 model) and they are getting pretty close 
I highlighted all the similarities here:
Tbf I can’t tell which is which at first glance
The main giveaway that lovable is on the right is the lucide sidebar icon in the top left of the header which is a common shadcn pattern. Then the bb-code issues in the Bubble sidebar!
Lovable’s is def smoother. Also the key is the wiring. Bubble’s is not wired to data - thats why its “-” and not zero. If nothing is wired to any data, I’d much rather use Elemium’s or Nocodable’s components - assuming I was pretty new to Bubble and didn’t have my own components.
I think p’ship with Claude is nice but what does it do if there’s no actual wiring being generated? MAkes a barebone UI scaffolding a bit nicer, sure but to what end?
Tomorrow I’ll publish a case study on Bubs which demonstrates what AI building a bubble app can and should do…
Standing about a foot away from the screen, I would agree. I would also say the biggest issue is the lack of common design patterns in native Bubble elements. For example, this clickable sidebar that collapses (highlighted below). Things like this are hacky to build and require a lot of duct tape that a model like Sonnet 4.6 isn’t going to be able to replicate. Or something like a simple selector with 3 options. Extremely easy to do with regular AI, but impossible with Bubble’s AI because there’s no native “selector” element. You have to use a bunch of groups and conditionals to replicate that behavior.
Things the AI can’t do because Bubble itself can’t do them:
Collapsible Sidebar (would take at least an hour in Bubble):
Selector (simple common design pattern):
Toggle:
The list goes on. Drawers that open from the side of the screen, etc. Even for elements that currently exist, some are very weird and look 10 years old (the Slider element’s handlebars, for example). We only recently got a functional Table element, and I think before that the most recent one was the Multidropdown element. I’m not even sure the AI can create pages with these two elements, let alone the things I just discussed.
Essentially, the AI is only as strong as the underlying native elements, which are themselves inadequate. If a new user is going to come across all these limitations, they’re going to pick the slop builder over Bubble ever time. It ultimately doesn’t matter if Bubble can replicate a 2013-looking UI without the expected functionality of a modern-day app.
With bubble AI maybe, I’m not sure, as I have not tested it. I can do it in about 1 minute with a dynamic width plugin, not sure with no plugin but possibly 10 minutes maximum.
Are you struggling with this collapsible design at the moment?
Hi Paulo, I just reached out would love to learn more about this on a call! Curious if this was with our AI app builder or when prompting the agent? Sounds like the AI App builder but wanted to clarify.
