Bubble.io vs AI Coding (bolt.new, V0, Cursor AI, Replit, Claude AI)

[vovahumnytskiy]

Hello, guys.

Just curious if anybody is considering completely switching from using Bubble.io to AI coding.

What are your thoughts on this?

Do you think people will use Bubble less and less over the years?

How are you preparing for this shift? I’m especially interested to hear opinions from those who do Bubble for a living.

I just saw a Bolt.new tutorial, and it seems like a good option already available on the market that might be better than Bubble (haven’t used it yet, just an assumption from the first glance).

What might be the disadvantages of AI coding over Bubble?

[sem]

We are shifting more and more away from Bubble. Our backend is now Supabase and all the customization are fully done by AI. So things like subscription management, dynamic distance calculations between users combine filters based on various filter sets, analytics and much more. All is done by AI and implemented in mostly Postgres and some Deno

At the moment we do a test with building various Saas tools. We use Saas tools for landingpages, email marketing, blog and more. Now we try to setup an architecture and let AI build it . So far impressive, but it is hard at stages to keep oversight and steer AI correctly.

But, it is amazing how fast AI can solve difficulty or error prone 200 lines scripts in seconds.

Tip
Stick to mature well known bits and pieces
AI will write better JavaScript code than 95% of the JavaScript coders out there. Same for Postgres. Ask it about Bubble and the response will be Mediocre.

[vovahumnytskiy]

@sem

Tip
Stick to mature well known bits and pieces
AI will write better JavaScript code than 95% of the JavaScript coders out there. Same for Postgres. Ask it about Bubble and the response will be Mediocre.

So, you mean AI is not good at building apps using Bubble and is good with JS/Postgres? If AI can code an interface with JS x5 times faster than a Bubble developer can do the same interface, doesn’t that make Bubble obsolete? Aren’t we there yet?

[sem]

If you are a developer I would say we are there already. But if you are not a full stack developer there are some things which are still hard to get right.

• AI does hallucinate now and then so you need to glance over the code to catch it
• AI is getting it hard when there are too many rows in a single script doing too many different things
• Although it can give you good architecture advise, it might be not precisely what you want or need
• Advise is sometimes too general. For instance, advising against SQLite because of “concurrent write issues” but when you know it can do easily 50.000 writes a second you know that 99.9999% of all websites in the world will never ever need 50.000 writes a second.
• It finds it difficult when it needs to chain too many languages in a single prompt. So suppose you want to handle authorization with Keycloak, db with Postgres, front end with Vue, server with node and styling with tailwind it will have a hard time when you say “I want the user to be able to login with a magic link via pressing a button in a pop-up” and expecting it will setup all pieces.
• It has a hard time to change code in a chain of a fully functional product. So suppose you want to add or change a feature, uploading 100.000 lines of code spread over various programming languages will not work. So you have to split and categorize it.

But if you can program or know just enough about programming it is already amazing. It simply cranks out code faster than you can type. So even if you would be a bad ass programmer that doesn’t have to think about how to solve a difficult thing with code, you will still be typing slower.

Also very interesting how it is able to follow instructions like “I need you to analyze this piece of code, minimize the number of rows with 20% and improve performance by 10%”

[boston85719]

That little red ERROR is why I personally would not be using an AI app builder like Bolt. If I knew how to code and could spot those errors and fix those errors, I’d consider it, but since I am a non-technical person, I do not anticipate an AI app builder to do everything I would need it to do, such as not produce code with errors.

[vovahumnytskiy]

Yeah, but I believe there is a good chunk of users who have used Bubble before to build their products but have never touched the Bubble editor. I mean, they build no-code teams who build for them, not build themselves. Then, they might consider using AI coding if it’s faster to ship features with it.

[sem]

These errors are easy to solve for AI but you are right. If you do not know how to code and you want to make something complex you cannot do it with AI. But to be honest, nobody without any Bubble experience is able to build complex things in a few weeks with Bubble.

So there is the question, do you want to learn just enough about a few programming languages to use AI with vanilla code or do you spend time in a platform.

Interesting times!

[fabrice.ariesdrinks]

it’s very promising especially on the windsurf side, I’m testing an application coded using windsurf AI for a period of two months as a beverage wholesaler using my ERP developed on bubble.io. I reproduced the ERP (beverage wholesaler) created in 6 months on bubble in a single week on bolt.new with very good instructions. Bolt has very big limitations so I imported the source code into windsurf to improve it and the latter has solved 100% of the problems related to bolt’s limitations, it’s so exciting to wait for the problems I’ll encounter on the road but so far everything has been solved and it’s going in the right direction

if you were able to learn bubble.io, you would be able to learn these tools more easily, you just need to know what you want, what you are creating and what you are modifying

[salemmo409]

As long as the development costs using Bubble and NoCode are significantly cheaper than the development costs using AI, there will still be a market for NoCode developers even if the speed of building applications using AI becomes faster than NoCode, and the NoCode will remain the preferred choice for those who want to build their own applications because learning and practicing coding is painful while learning and practicing NoCode is fun.

[fede.bubble]

can’t speak for the whole team here but my 2 cents: Bubble is still an incredibly powerful ecosystem to build (not just the editor but availability of agencies, plugins, templates, community, support, etc). However, AI tools are coming up with a lot of cool features that promise quality and speed. At least superficially, these tools seem to be satisfying some user needs pretty well.

I said it before and I’ll say it again: no-code doesn’t mean no-plan. And I believe the same is true for AI. An inexperienced user with ai tools might get 70% of the way there and then get stuck (same is true for Bubble btw). But I think Bubble is better positioned to solve that last 30% than the current combination of disaggregated AI tools.
I’m guessing most users today are wondering how is Bubble solving that? which is why us rolling out Native Mobile, AI assistants, platform stability and security improvements, workflow tab work, Editor updates, and a ton of other things comes into play.

[ihsanzainal84]

I’ve said this before and I’ll say this again: Until AI can solve issues that occur in a live app without my intervention, it’s still just a tool. Building an app is one thing. Troubleshooting and debugging is a whole other shitstorm.

I build my plugins with AI and it’s great! But I can also troubleshoot along with AI when it messes up, because I can read and understand code. So in my experience, there is still a cost of entry even with AI.

We also need to remember that Bubble is a full stack solution. With Bubble I don’t have to worry about anything other than optimizing how I build my Bubble app.

When something goes wrong in the backend I can blame Bubble instead of shitting my pants going through massive error logs to then fix issues from multiple cloud services while managing angry clients and users.

[code-escapee]

@boston85719

That little red ERROR is why I personally would not be using an AI app builder like Bolt. If I knew how to code and could spot those errors and fix those errors, I’d consider it, but since I am a non-technical person, I do not anticipate an AI app builder to do everything I would need it to do, such as not produce code with errors.

@sem

These errors are easy to solve for AI but you are right. If you do not know how to code and you want to make something complex you cannot do it with AI.

@boston85719 & @sem , you guys are correct that TODAY you can’t use AI to build a complex app without knowledge of how to code and spot and fix errors. But considering that the Moore’s Law lifecycle for AI is only 6 months , in 2025 there will surely be a few solutions released that function as a wrapper/GUI around the coding components of these amazing AI app builders to allow us non-coders to read and modify the code.

---

@salemmo409

As long as the development costs using Bubble and NoCode are significantly cheaper than the development costs using AI, there will still be a market for NoCode developers even if the speed of building applications using AI becomes faster than NoCode,

Which AI dev costs are you referring to? For a robust app, the costs of Windsurf or Cursor are ~80%+ cheaper than no-code solutions and like 95%+ cheaper than the pricey no-code solutions like Bubble.

[salemmo409]

@code-escapee

Which AI dev costs are you referring to?

I am referring to the costs of hiring developers, there will still be a market for Bubble and Nocode developers if the costs of hiring them are less than the costs of hiring coders who use AI.

The market for NoCode tools is likely to continue because it is preferred by those who want to build their own apps. However, the continued existence of a market for NoCode tools does not mean the continued existence of a market for NoCode developers. The market for NoCode developers depends on their speed advantage and the low cost of hiring them. The market for NoCode developers can continue if they lose the speed advantage if the cost of hiring them remains lower.

[pipeleteli]

I think the challenges with AI applications will become apparent when production applications need to handle technical debt or make changes that affect existing functions without disrupting them. The risk I see in relying too much on AI to build apps is that you end up breaking existing features by trying to make new things compatible with old things, or creating “junk” code and redundant data schemas to avoid breaking things that are already working. creating a huge amount of technical debt that will eventually impact the performance, cost and maintainability of your application.

This is where I think Low Code has the superior advantage, in a way low code keeps your app split into “isolated” modules and you can use the power of AI to create UI, logic and custom code faster without fear of breaking your entire system.

Where I feel Bubble is falling behind is that it is trying too hard to keep everything “No Code” when there are kids creating functional code using only their natural language. If Bubble would work on improving the plugin editor to make it easier to connect code co-pilots and allow us to work with JSON/Objects freely in the editor, it would already be a big improvement. Tools like FlutterFlow are already evolving and balancing AI+NoCode, but because the mindset of these tools is to keep the best of both worlds instead of trying to get rid of one or the other…

So, I don’t think Bubble will get obsolete, but definitely will be outplayed by other Low/No Code tools if they can’t keep up, BUT you will still be able to create magnificent apps with Bubble without having to deal with techy issues, so it is up to you if you want to deal with more technical problems or not by yourself, because good luck trying to find someone that will help you when your messy and huge app created with AI its broken and you can’t improve it anymore just using prompts.

[munaeemmmm]

We have moved out completely from bubble using supabase and next js. possibilities are endless

[sem]

After 6 months of running a fast amount of coding tests and testing code in production I can tell you that when you know how to prompt AI and how to structure your workflow and way of thinking you get very high quality code in minutes.

It is true that AI has a hard time to take garbage in and turn lovely code out. So does a programmer.

People in general have absolutely no idea what is coming and therefore it will hit them hard. Very hard.

I was also involved in tests of AI against lawyers and doctors. AI had better advise than all of them. And did it in minutes, not days or even weeks.

I’m confident that I can write any Bubble project out here in 80% of cases within a few weeks with AI. With plain vanilla code. Super fast runtime and hosting plan of a few dollars a month to support thousands of users. I can because I understand a lot about coding and have fast experience with AI. I think it needs another 6 months to get to a point that less technical people can do the same. If you know how to prompt and ask the right questions.

For instance, need a responsive landing page with form that validates all fields and calls an api to send the data to and includes all kinds of fancy interactive elements? A single prompt and you will have it in under a minute.

Need to parse 20.000 messages and pull al kinds of statistics from it? Prompt it and you will have the script in under a minute.

Need an api call to strip? Drop the link to the stripe docs, link to the samples, prompt and you have your working api call in under a minute.

Need to have a query that analysis which tables and columns are used the most, do not have an index, find the best indexing strategy, change the config file, set the indexes and run some tests? Prompt it and you will have it in under a minute.

Have existing code that is running into an error? Prompt for a strategy to investigate and solve, do the steps and solve it in minutes.

Have lots of data and want to have fancy d3 graphs of it? Prompt and get it in under a minute.

It are exiting times!

[pipeleteli]

Fair, but don’t get me wrong, I’m not saying you can’t create great apps using AI, on the contrary, I agree with you that it is quickly getting to the point that it is a “must” to use agents and copilots to keep up with the competition.

I use Cursor everyday to create integrations, snippets, widgets, components, you name it, but I know what I’m doing, I know how to “prompt” my way through and that’s because I know how software works, I can easilly tell when AI is creating redundant classes or when I should be creating an exportable function rather than hardcode it into some file. But you can’t expect from people that haven’t developed an app in their lives to get “AI prompting” right without understanding how code works, actually, I think that’s dangerous! Imagine by mistake you expose your private API keys just because you don’t know it has to be private. It isn’t just about being productive, is about security and scalability too.

That’s where using Low Code tools like WeWeb, Xano, Toddle, vs hosting your code yourself has its benefits, you don’t need to spend lot of hours handling dependency confflicts, setting load balancers, add cloudfare by yourself if you want the extra security agains DDOS attacks or scrappers, etc. you just focus on building and nothing else. And these tools are evolving too! Flutterflow has a lot of AI features from converting draws/prompts to UI components, to prompted logic builders, plus you can export your code whenever you want, so you can take it to an ai agent at any point. WeWeb is also making amazing progress in this direction.

If you want to dedicate your life to create software I would encourage you to learn to code a 100%, but if you want to solve specific problems, create internal tools and don’t have the time to learn more techy stuff, I would consider getting into LowCode tools that leverage AI+No Code first, but tools that allow you to export your code or that are at least opensource.

At the end, all is about knowing what you are doing, even if the AI is doing the heavy lifting part of the job.

[tanujarajan91]

@pipeleteli Which are the best tools you’d recommend that are “LowCode tools that leverage AI+No”? Is it FlutterFlow and WeWork like you mentioned?Thank you for your insight btw! I previously learned Python, but am not adept at it since I don’t use it for my work/life. Other than that, I don’t /can’t code. What would you recommend?

[sem]

Any tool out there is having it hard to connect the dots when you expand your code. That’s because it looses context.

We run tests now with the platform we have build. We build a platform with vanilla code that can be extended by vanilla code or AI but it does let you drag and drop functionality. So things like api calls, payment, CMS, crm, SEO, blog, authentication, magic link, transactional and email marketing it all works out of the box with full white label support. This way, context for 80% is fixed and we only need to train AI with this fixed fundament. On top of it you can build whatever you want.

We now try to let multiple AI models code themselves. They use each other to brainstorm in requirements, write architecture using the fundament structure, propose code, implement the code, run tests, adjust in case of errors etc.

We will get there but it does require lots of tests, rewrite of code and therefore still 6-12 months to be fully ready.

[sarsasparillaj]

It seems like it might be more doable if you code properly, breaking things into classes, objects, functions and keep each in a separate file so when you feed it into AI, you are only messing with a small part of your code at a time.

I was a regular developer years back, then came to Bubble to get something out quickly.

Can any of these AI tools hold a candle to Bubble when it comes to UI/UX design and tweaking real time? If so which ones? I’m curious to see them.

AI makes does of sense on the back end though and that’s where bubble is a bit weak. You often have to jump through crazy hoops to do simple things because they tried to keep it “simple” here