Found something that fits this topic (and no this isn’t a fake)
Looks like maybe ai wants to be more nocode
I actually predicted this would happen 1 year ago:
It’s similar to this concept, but in a more particular domain:
After using bubble and also vibe coding excessively recently for real world projects (including buildprint for “bubble vibe building” the last few days) I can say with certainty that there is room for BOTH types of building, infact they go hand in hand & compliment each other well. Each is quicker for certain things , and slower for others. Better for some things, worse for others.
Vibe + visual control within a lightly walled garden IS the future, I’m convinced.
Bubble is in a good place now it just needs to ensure it’s fundamentals are rock solid… performance, reliability, SEO, scalability…and it needs to double down on what it already does brilliantly. With that strong CORE, the AI doesn’t replace bubble - it simply makes it more powerful & useful. 
They said the same thing about news papers.
Which thing?
I completely agree with Bubble’s potential and vision.
However, for Bubble to become truly mainstream, the usability of the ‘Figma to Bubble’ plugin needs a massive overhaul.
If not, they need to offer an ‘export to Bubble’ feature directly from Google Stitch—just like Stitch allows exporting to Figma—or they need to provide a reliable ‘HTML to Bubble’ function.
Let me break this down based on my own experience.
- The vast majority of people flocking to Bubble don’t know the first thing about code, and we certainly don’t know what JavaScript is. I’d wager most are just everyday people, not “programmers.” (If that’s not the case, Bubble has no future. Wouldn’t programmers just find Codex or Claude Code more convenient and efficient?)
Ordinary folks usually find themselves wondering, “How can I build this?” and end up at Bubble because they hear it’s an easy, fast, and fully integrated tool.
- That’s because there is simply no other no-code service that integrates server, database, web hosting, and design quite like Bubble does.
Anyway, up to this point, it’s pretty similar to how someone might stumble upon Lovable, base44, or Figma Make.
As someone who chose Bubble, I came in full of high hopes. Since they emphasize being a “full-stack no-code tool,” I thought, “Ah, I really can build absolutely anything.”
- But when it actually comes time to design a screen, you realize you don’t know much about design, and you just want the tool to handle it for you. Even if you try your best to draw it, you’re not a design expert, so it doesn’t look great.
Then it hits you: rather than learning to design from scratch, it’s faster, easier, and yields a much better-looking screen through intuitive communication if you just use Stitch or Lovable.
- That’s how you realize there’s absolutely no reason to draw the UI directly in the Bubble editor in the first place.
This is exactly where a lot of people will drop off. - I tried asking Bubble AI to draw it, and it clearly falls short of Google Stitch or Lovable.
- Fortunately, though, Bubble provided a workaround.
I designed a beautiful screen with Google Stitch, pasted it into Figma using Stitch’s ‘export to Figma’ feature, and then brought it into Bubble using the ‘Figma to Bubble’ plugin.
Thanks to that, I thought, “Great. Now I just need to bring it into Bubble,” but…
The actual process of importing it is pure hell!
The ‘Figma to Bubble’ plugin is a remarkable concept. But once you actually import something, you end up feeling like you wasted your whole day just tweaking widths, heights, and alignments.
From the very beginning, Bubble was a service for “designing” screens, not “coding” them.
I believe the “no-code” Bubble envisions means “programming via WYSIWYG design.”
But when ordinary users come to Bubble with high expectations, they easily give up once they realize the design part is just tedious manual labor.
If Bubble doesn’t improve the UI and enhance the features of ‘Figma to Bubble,’ and if it can’t offer usability that surpasses the Nocodable extension—which provides batch style editing and component saving features…
- Honestly, for a regular user with no coding knowledge, choosing between learning Bubble or learning to code becomes a toss-up. After all, tools that make coding easier are constantly evolving.
Based on my experience, I strongly argue that Bubble needs to allow regular users with no coding knowledge to simply copy and paste the screen itself.
It would be wonderful if we could use an ‘export to Bubble’ feature directly from Google Stitch, but for now, I propose these two solutions:
- Upgrade ‘Figma to Bubble’
- It needs to be perfectly compatible to the point where tweaking width, height, and alignment is completely unnecessary. Just having this would double the productivity of no-code Bubble users. I never want to experience wasting over three hours adjusting a single pasted screen ever again.
- Introduce an ‘HTML to Bubble’ feature or plugin.
- If that’s not possible, another way would be to draw it indirectly using an AI tool that inherently understands Bubble’s logic, something like Buildprint.
If they can somehow make it possible to import an entire screen with a single click, users coming to Bubble will have no reason to turn their eyes to traditional(?) AI coding tools like Claude or Antigravity.
- How long would it take for an ordinary person to gain the ability to instruct an AI on servers, hosting, database architecture, security, and design—or to review the AI’s output and control everything? It’s impossible.
For a regular guy like me, knowing how to handle domain/hosting, page design, and my required workflows is more than enough to build my own service. At least, if I’m using Bubble.
- However, page design is the biggest factor making people turn away from it.
- Compared to code, the workflow is so intuitive and this is irreplaceable. It’s Bubble’s greatest strength. How are you supposed to write logic and create events when you don’t even know JavaScript syntax? *Even if it is easier than coding, some will give up while learning how to use Bubble’s unique workflows (events, actions), but we’ll leave that aside for now.
- Domain/hosting is also incredibly simple. The only problem is the design.
Bubble is leading the true “no-code” movement, but it’s being chased hard by no-code tools backed by coding.
I think it’s crucial now that Bubble let us directly use “designs created by code” right inside Bubble. @fede.bubble @emmanuel @josh
did you know https://buildprint.ai will convert code/figma to Bubble and build it for you?
This is a very vibe coded design for test purposes:
Spoiler - that’s the one-shot screenshot from Bubble, not the original design!
Agree with this fully.
The cons and expertise needed were at the core of why I was fully sticking with Bubble. However, a few things that have changed recently have pushed me in the other direction.
-
It’s generally getting better and a surprising rate(this isn’t nearly enough, but the improvements have been impressive.)
-
Competence with AI coding is no longer the vibe-based nonsense of “learning to be an expert prompt engineer”. It’s now about creating systems of rules, having multiple agents confined to those rules, and setting them to operate within context windows/token limits that aren’t subject to the degradation of what people thought they could do with “read my whole codebase for me and make me a new big feature”
-
There are increasingly helpful guides, repos, and such created by software engineers who COULD code the stuff on their own, but are leveraging these tools to make the AI integrated into IDEs, to run intelligently at the proper direction of the user, with additional automation of essentials like unit tests, CDI, language-specific code review, security review, etc.
-
The drastic improvement of the models also allows you to learn concepts you need to understand your code, by asking these AI tools themselves. The time from confusion to answer 10-100x faster when ‘chatting’ with these AI tools in the context of your code/repo.. as opposed to finding something close-ish across 8 StackOverflow posts that don’t fully fit your use case. Now, it can explain parts of your exact code with greater context, it’s no longer just pasting lines of code without context in ChatGPT and having it hallucinate half the time while you pretend its reccomendation is going to work in your vibe-coded app.
So yeah, I was getting annoyed with the hype/propaganda posts of “I switched out of Bubble to AI and here’s why you should too” (almost always generated by a ChatGPT itself). Nearly as annoyed as I was with the vibe-coded slop apps that people were claiming were full-fledged, secure applications better than apps built with intention and skill by bubble devs, all claimed to be the future with “the new skill of prompt engineering” to get it in one shot.
It’s not prompt engineering now, it’s complex agential setups, and having the knowhow to utilize these tools in development workflows.
While “copy to Bubble” (or “fix vibe-coded app” for better marketing) is an amazing top of the funnel tool to get new users into the ecosystem, that’s not Bubble’s main problem. It’s churn. That has a separate set of solutions.
Honestly, I think people are looking at this the wrong way.
There is no “best tool”.
Each tool works for a specific need and a specific type of builder.
I’ve built large systems using AI (vibe coding), and it’s definitely not bad — actually it’s very powerful.
But the difference is: I know what I’m doing, I know what I need, and I can debug and fix things when AI messes up.
Without that, AI will just slow you down.
At the same time, Bubble is still very powerful — especially for understanding things visually.
Workflows, data flow, logic… it teaches you programming in a different way.
So it’s not AI vs Bubble.
AI gives you speed and flexibility.
Bubble gives you clarity and structured thinking.
Both are useful.
Both have limitations.
At the end, it’s not about the tool —
it’s about the person using it.
I really hope everyone gets to try it out as soon as possible, George. The features your Buildprint offers solve the very issues I’ve been talking about in one fell swoop.
Once that happens, Bubble will easily become a faster, more polished, and more secure no-code solution than any vibe-coding tool out there today.
From everything they’ve done the last 2 years, especially how they are approaching AI - all style and little substance and only willing to pay for the token for NEW apps - it seems their goals are all top of the funnel (prob looking for another round?).
Bubble’s retention has cratered the last 12 months but it doesn’t mean that thats their priority…
My point is they have absolutely nothing to gain from increasing their top of funnel when churn is the main issue. It’s like pouring a bucket of sand through your hands.
and my point was that they clearly are intent on increasing their top of the funnel despite not retaining 99.8% of the apps. If the goal isn’t to collect sand then you’re ok with it going right thru your hands.
Where did you find that metric? In an AMA Emmanuel said retention is going up or maybe it’s the more important metric of users becoming paid users.
Do you know what the percentage is?
there is no evidence to back this up btw, apps that onboard using Bubble AI have a higher probability of deploying a live app on a paid plan than apps that do not as well
Where’s the evidence for that? That seems counter-intuitive especially given how toothless the AI is, it’s usually much easier to start from scratch.
bubble’s own metrics
i was surprised too
whether those apps are more likely to be subscribed in twelve months after deployment than non-AI built apps is a different question
Exactly, where is your evidence to say it cratered?
Do you know what the rate was before AI and is after AI? Or was the comment just a hypothesized statement built on sentiment?
Emmanuel answered the question in an AMA…it’s going up.
sounds like some creative bean counting but what’s higher even mean? 3.2% for AI generated vs 3.15%? and yes even 3 months after I wonder if its more likely to be subscribed.
It’s also NOT a clear distinction between AI and not. I often start with AI generated app (in the rare situation I’m starting a app in Bubble) as a very rough scaffolding and because I’m optimistic it will have become more helpful since the last time I tried.
But the AI generated parts are completely changed and not one dataset has majority AI built fields or ANY of the original names. so was it really AI generated?
and more critically, a clone of a AI generated app is still deemed as a AI generated app! (even though the Agent is no longer available).
So are truly AI-generated apps truly higher converting with any significant retention? I’d bet no and until actual specific data that isn’t part of a face saving campaign proves otherwise…