Even AI models didn’t do this well* until Codex 5.3. Difference is:
- those models have tons of training data on typical programming languages, Bubble’s JSON language isn’t in the training corpus
- there’s very consistent/reliable branching/merging via Github whereas Bubble’s undo/redo function is terrible and there’s no proper changelog ( In your experience, how often is the undo/redo button buggy? )
- I don’t see why you would let an AI ever edit the entire JSON file when the beauty of Bubble is its meta-level “type safety,” meaning the lego blocks either fit or they don’t. Cutting and splicing into the entire structure seems like a really dangerous idea when you can instead reliably generate perfect lego blocks every time and then allow the user to place them in the right context.
I have faith, I just think they will take a long time to eventually come to my conclusion (Monthly Community Update - January 2026 - #16 by randomanon). It’s going to be the “Beta Expression Composer” all over again. It’s funny because they wasted hundreds (thousands?) of man hours on something that wasn’t technically feasible and might be doing the same thing all over again.
*and arguably still don’t depending on what your standards of “well” are