For the last few years, we’ve been using Bubble for our production multi-tenant software. It’s a vertical CMS with project management and knowledge management features (very file and records based).
While Bubble has some limitations, we believe our velocity has been faster than building with traditional code. At least over the last few years.
Now, with the advent of AI and the emergence of website building tools (e.g., Bolt, Cursor, Supabase, NextJS, Lovable, Windsurf, etc.), we’re not too sure. We’re considering refactoring our product using an AI enabled stack. Of course, this means we have to figure out infrastructure, etc.
In others’ experience with all the AI developments, are founders and agencies finding more velocity building with AI enabled stacks vs. Bubble?
Personally I use cursor, Ai does great at the initial build, however after so many iterations or losing its context window it gets bad, unless your building MVPs to launch quick with minimal iterations I wouldn’t use AI to fully build a production product the size your described your app is. In time of course AI will improve but right now i don’t think it’s there.
Do you have experience using cursor to help write code for maintaining and add features after launch? We believe it would excel at this as long as we have a developer to prompt and integrate the generated code into the application.
I’ve used supabase on a few projects… total game changer. If all you’ve known is the clunky, limited, slow, buggy and expensive bubble database then supabase is massive.
As for the front end. I’ve played around with lovable and weweb - transferred a few of my smaller bubble projects to weweb. Lovable is fine for MVP. WeWeb seems to be a good front end for supabase.
It is definitely slower to get to launch with both lovable and weweb than it is with bubble (perhaps that’s just my inexperience with the tools) but both are scalable, exportable and very affordable.
Google firebase studio was recently released… currently pretty garbage but if google gets into the lowcode space that’ll be really exciting.
I use Copilot in VS Code to build help me build my stack. I’m on it every day and AI coding assistants are not reliable enough to handle something like a CMS. It makes you think everything is fine and dandy but when something goes wrong…the time spent debugging AI is a horrible experience.
There’s nothing wrong with keeping most of your stack in Bubble if it is already working for you. Just integrate where you need to. For example, I’m currently upgrading and adding new features to my Bubble apps with my own tech stack (AI stuff, encryption etc.) all built with Cloudflare. They integrate to my stack using API calls to CF Workers which then handle the heavy lifting. I reap the benefits of both Bubble and AI coding assistants.
@sdharssi feel free to send me a DM for a bit more info.
We have build our whole project only with AI. Things like full multi language
Multi-tenant
Multi language email
Email template system
Web template system for web
Authentication
Authorization
Device management
Caching
Auto documentation
Detailed standardized logging helping developers and AI
And much more.
You have to build some things first in order to leverage AI well. We are now taking it one step up and building our own context system and multiple specialized agents. Trying to get to a stage where we have a natural flow from high level business wish to technical design, coding testing, deploying and writing manuals for non coding users. Most of this all we have in place but it is not connected yet.
Ultimate goal: you can prompt the system with a high level business wish, the system will start asking questions and giving directions if you have no clue until after a few minutes you have a test report on features delivered.
Thanks! Super helpful. We’d have to migrate 40+ tables plus update all our searches if we changed databases from Bubble to Supabase. But maybe it’s worth it? Alternatively, we’re thinking of leveraging Azure and the services they provide as part of Azure, to enable AI related use cases. Still early days but we’re making progress in complimenting Bubble with these tools.
Thanks for the feedback! We’re strongly considering the same, currently evaluating services in Azure to compliment our Bubble application. Planning on connecting view API.
Maybe a noob question, but how are you building in Cloudflare? If you have some documentation or training resources you can point to, that would be super helpful for our team.
Cloudflare’s documentation is top notch and easy to navigate. Quick links to their docs are around the user dashboard.
The gist of Cloudflare is that everything is centered around Workers. Need to call your DB? Just call a worker.
Need to Oauth, encrypt and then store to DB? All through a worker.
I code only for the worker, everything else is just settings or some wrangler (their CLI extension).
I don’t think the AI technology is there yet to build a production-ready app or something beyond a basic MVP. But, it’s only a matter of time before they are, like in the next 6-12 months.
I’ve been using Windsurf to customise open source frameworks like Crawl4AI (for scraping) and Browser-Use (for automated browsing) and connecting these to Bubble via an API.
However, beyond a certain level of complexity it starts getting confused and introducing changes in places it shouldn’t and productivty drops to zero!
So Bubble is still doing the basic stuff and the AI built app is doing more complex/customised task, albeit with a limited in scope.
I can see myself migrating more functionality onto the AI-built side as my confidence grows in it’s ability to handle things like authentication, databases, etc.
Personally, I’d hold off looking to refactor what you’ve got, until these platforms get a bit more mature. I don’t think the AI technology is there yet to build a production-ready app or something beyond a basic MVP. But, it’s only a matter of time before they are, like in the next 6-12 months.
AI isn’t the bottleneck. A monolithic codebase is. Build like a software engineer, not a prompt engineer, and you’ll never get stuck waiting for model upgrades.
I get where you’re coming from, but honestly, if you build your app modularly—the way any solid software engineering curriculum teaches you—you don’t hit the “AI context window” wall.
When your architecture breaks things into clean, composable modules (think: UI, business logic, data, AI integration, etc.), you’re never at the mercy of a single AI call or some massive context window limit. Each feature or workflow keeps its own state, and you only ever pipe in just enough context for the task at hand.
If your team has experienced devs, then that most makes sense for the long term. If the team is mainly no-coders and experienced Bubblers, then I don’t think the existing AI infra is there to get you to real mature products as a no-coder.
Agreed. The main problem with most vibe coders is that most of them are not programmers themselves so they lack understanding of how best to work within the limitations of coding with AI. I use Copilot in VS and make it a habit to not have long conversations. I now use AI only to write and review code blocks (cause I hate syntax) instead of letting it run amok.
It’s good practice to get the AI to write comments in the code. Comments in the code help when you start a fresh conversation to avoid hallucinations in large ones. It’s also very important to manually review the code ever so often.
That said, I never ever just give the AI a concept document and let it run wild. The horrors… I instead get it to work in steps: wireframing > scaffolding > review everything first > and then start putting code together. It’s important to micro-manage AI for complex projects.
For the services you are building outside Bubble using Cloudflare, do any require the processing of large amounts of data or files stored in your bubble app?
Our current AI use cases require processing and learning from large parts of our data and file sets. So far, we are finding that our hands are tied as we can’t leverage third party platforms very well due to the transaction costs involved in updating data and file sets in the third party platforms.
Yep, it’s one of the reasons I decided to stack Cloudflare. I like keeping base functions of my enterprise suite in Bubble as much as I can.
The tricky part was figuring how to get AI to work with a ton of different data from Bubble. So i decided to keep vectorized versions of the data I want AI to access, in CF Vectorize. You’ll be surprised at how cheap it is to integrate Cloudflare.
