C. I have a Bubble app that I want to remake with vibe coding so I can abandon the Bubble prison, but I’m also building with vibe coding now and keep hitting the same problems as many others such as running out of context at inopportune times, AI doing the exact opposite of what I ask it to, doing stuff I never asked about at all including deleting files or directories, saying it finished something when it really didn’t, overcomplicating the most simple instructions, etc. So I’d like to learn how to avoid these types of problems in general.

@Zubair Kakakhel Here is how I found this group and why I joined. I have a Bubble app which uses plugins but none that work exactly how I want. So my app has a ton of hard to manage workarounds. I started playing with vibe coding because I got the idea I could build my own plugin to work exactly how I want. Someday I still will. Along the way I decided to try building a chrome plugin. This is where I started having some problems. Without too much detail, I’ve learned all the things mentioned already: small chunks, PRD, tech spec, implementation plan, direct instructions but trying to not fill the context, etc. I do all of that and yet the AI still does the opposite of the instructions. At this point I saw something from you, maybe a YouTube video or a post on the Bubble forum. I can’t remember. But it was about a full app build so it caught my attention. And here I am!

I’d love a video on that. I approach it by ignoring it! 😐

What was the running cost for Replicate to connect to nano banana? Will the upcoming api be a better pricing? Also, what is the approximate railway cost to host this? I’m trying to see what might need to charge to reach the break even point for a SaaS like this. Thanks!

I’m using Claude agents. They are a game changer for me. I’m working mostly in Claude Code inside Cursor. I was having a problem where AI would do stuff it wasn’t asked to do even though it had specific instructions about what to do, the ‘plan’ looked clean, and there were general instructions in place in my Claude.md file. And it added complications in a way that I had to go back and clean up the mess every time. With clause agents I am able to have even more focused instructions. I have an agent to do the coding I asked for which repeats the specific stuff to do and how to do it (emphasizing to not over engineer) and then I have an agent that runs after new code is created to validate that it followed the instructions, didn’t do stuff it wasn’t asked to do, and validate the code that was created for errors and for compliance with the systems it is building for. With each agent having its own context, it is able to run for a very long time unattended and give me mostly clean code at the other end. It’s way better than sitting there troubleshooting! Next I’m going to work on learning hooks for a more reliable way to trigger the agents. Right now they usually run when they are supposed to, but not all the time. Do I have to prompt specifically to use them.

Hi! I’m Ken and I have been working in Bubble for a few years now. I have a productivity app out that I am trying to grow. Also, starting in April I began vibe coding. Fast forward and now I’m mainly using Cursor with Claude Code. I’m using it to build a few Chrome extensions. But I also want to get proficient enough to move my Bubble app to a full stack and move away from Bubble. Glad to be here!