As AI-assisted coding started taking over the software development industry, teams maintaining Bubble applications were starting to feel like they were falling behind.
While traditional engineering teams were accelerating with AI tools, many Bubble teams were still dealing with the same manual workflows around planning, building, testing, and debugging.
@randomanon I’m referring to the BuildPrint agent in edit mode. We tried an early version while building a new feature in a fairly large Bubble app. It didn’t end up implementing the feature exactly the way we intended, but the experience was still promising. I think it can be really useful for other use cases.
We gave it a spin a few weeks ago, but we’ve really started leaning into it this week and it’s awesome. We found the best workflow is starting in Plan Mode to get the logic right, then moving to Build, and making sure to create a dedicated branch for each new feature or update. It’s made the whole process much cleaner.
It’s wild to see a Bubble Renaissance happening in real-time. I remember one of the initial @georgecollier posts on X re: BP’s editing capabilities and thinking “this is gonna be Bubble 2.0.” George and the BP team haven’t disappointed!
Edit: George’s stated initial goal was to “get people excited about Bubble again”. So cool.
I was playing around with an idea to take a Figma design and have it reproduced in Bubble.
It is 100% possible.
I tested it on a button and a couple of other small things, and it worked fine.
I just don’t have the time or energy to follow up and make it for a whole page, etc. Usually, when I do get free time, I’m either wasting my time on some forum or just vegetating lol.
But it can be done.
Would also think that if BP could examine workflows, etc., and give an idea of the work unit costs, that would be awesome. That also could be done… but again, it takes time.
You’re on a roll with BP, and I’m quite sure it will only get better and better
We typically switch between Sonnet, Opus, and GPT-5.5 depending on the task.
Here’s the workflow that’s been working well for us:
Start in Plan Mode and create a single branch, whether you’re fixing a bug or building a new feature.
Before writing code, ask the AI to understand the problem or explain how the feature should be implemented.
Ask it to generate an implementation plan.
Review and refine the plan before building anything.
Once the plan looks solid, ask it to create a new branch and start implementing.
For new features, build them as reusable components or on separate pages whenever possible.
If the task involves UI, have the AI design it in HTML first. If you’re using Claude Design for design, export the design and ask BuildPrint to implement it. Always tell it to reuse your existing styles (if you have one) so everything stays consistent.
Before merging, run a Review. It almost always catches something useful or suggests a better approach.
When the implementation is complete, ask it to generate tests and documentation before wrapping up.
A little planning upfront saves a lot of debugging later.