Hello all, I have some background with creating AI agents (and plugin building) and got this idea to create an AI web app that would build Bubble plugins by user prompting.
I am now working on a prototype and I am curious - would anyone here actually be interested in using that? And what would be the plugin you would create with it?
The main challenge in this case is that Bubble does not expose an API for sending code to the plugin editor. The most tedious aspect of plugin development is setting up properties rather than writing the actual code, managing states, etc. Therefore, your tool would need to handle this setup process as well. Consider developing an MVP (Minimum Viable Product) and potentially pursuing a partnership with Bubble directly, as this could help address these limitations.
Iāve been hacking away on Tiny demo: Pled CLI tool for plugin development - #17 by rico.trevisan .
I can get the LLM to create changes, push changes, open a browser to check the test app, and then iterate on it.
Quite a nice workflow.
If you can get the AI to spit out correct code first time, people will use it. If you just add a layer between existing LLM like ChatGPT-5 and code is wrong on first try 99% of time, people will likely not use it. Key is correctly built off one prompt.
How did you build Data Jedi without having a technical background?
I graduated from University with a Finance Major and Psychology Minor. I worked in Real Estate, Restaurants, as an ESL Instructor, Owned a Backpacker Hostel and Bar, failed attempts at tour company, a few other business that are profitable and Iāve been building on Bubble for 7.5 yearsā¦so my background is not technical from the perspective of āwhen I started building on bubbleā I had no background in tech to speak of. So now, after 7.5 years of building apps, Iām technically minded, still retain all my original abilities of thinking logically, and can combine that with financial/entrepreneurial/psychology background which just helps me have a different perspective on the business side of app development than some other developers whose backgrounds are in tech. It allows me to help my clients in building the technical power behind their business (the app) and the business strategy to drive their business.
thats pretty cool and I guess every plugin developer will want something like that
Thanks all for your thoughts on this. The setup the custom fields in the plugin editor is obviously a challenge. Yes, I suppose reverse engineering Bubbleās api or partnering with then is the way to go.
Any more thought on what plugins would users like you create?
My point was that you clearly didnāt do it in āone shotā using chatGPT or whatever other LLMs helped you vibe code your plugin.
Meaning, thatās an unfair benchmark.
(FYI: āTechnicalā in the startup/software world typically refers to programming.)
What causes you to come up with that assumption?
That is how I used it
You have to understand something. People pay for problems to be solved. So, the response I gave was under the impression that OP would try to monetize the tool he would build.
OP stated they have a background in creating AI agents and plugin building, which would imply they have more experience than most people when it comes to using AI to build a plugin and so could solve a problem that other people have, which is AI doesnāt get code right the first time.
Now, if the OP is not able to make something that is materially different than somebody just going onto any LLM and just vibe coding themselves, than they do not have a monetizable product, because who is going to pay for a product that is just a different UI accessing an LLM via API adding some latency and financial overhead to the whole AI building a plugin process. If all the OP is doing is creating a new UI layer and running API calls to some LLM, but has done nothing to solve a problem (the main one is to get the single prompt to create a polished working plugin), then nobody is going to pay him.
But, if the OP is able to solve the problem, and with their background in creating AI agents, they really solve the problem and make it so the single prompt results in a working plugin, then they have a product people will pay for. Otherwise, people would be better off to just go directly to either ChatGPT or some other tool and vibe code it themselves.
You coded the plugin without the assistance of AI?
Do you know what a CLI is?
Iād like to build plugins that integrate Bubble with AI agents. There are plugins for VAPI but there are many services such as Bland.ai which donāt have any plugins on Bubble marketplace
It is the thing in stripe that makes it more difficult than it needs to be to initialize webhooks
CLIs deserve love 
They may be intimidating at first but you will learn to love them as time passes. CLIs are the backbone of programming.
Haha this is one of the craziest things about Stripe to me. How is there no ability to send a test webhook from the web dashboard?
Yes, I suppose it is a must-have to make the plugin vibe-coding work better than OpenAI wrapper. For me benchmark is a Cursor/Claude Code with well written rules for building plugins, implementation examples, some helper functions and tools that accelerates the development similar to what @rico.trevisan is developing.
For me coding a first plugin was a nightmare but each other plugin I have done with more ease because I gradually managed to write the plugins type safe, made the Bubble plugin editor syncing process a lot faster (by some scripts), wrote some Cursor rules. Now I can quite fast create plugins in Cursor mostly by prompting and Iād like to expose some similar experience to other devs.
