I’ve only just started experimenting with Wappler’s AI Builder tools, and while the potential is clearly there, I’m running into a number of blockers and could use some clarification.
Context
-
I’m running multiple Node.js projects in Docker containers
-
Using GPT-5.1 / GPT-5.2
-
Working primarily with Server Actions alongside normal page/UI creation
What’s working reasonably well
-
Creating pages and UI elements via AI generally works
-
Basic UX/layout generation has had some success
Where I’m consistently hitting problems
Whenever I try to create or modify Server Actions, I almost always hit one of the following issues:
-
“The server action JSON file cannot be accessed”
-
“The server action cannot be edited”
(often because it’s either the focused file or not the focused file)
-
“This file is not in the expected path”
-
“The JSON schema does not match the expected schema”
-
File not found / file not accessible
-
AI failing because it can’t locate files at expected paths
In several cases, the AI has told me:
-
Server Actions cannot be created via AI
-
Server Actions can be created, but not edited
-
Sometimes it works anyway — but inconsistently
This often results in me going in circles:
-
Opening and closing files
-
Changing focus
-
Retrying prompts
-
Only to end up back at the same file-access or schema error
Other issues
-
I’ve also hit situations where the AI appears to enter an infinite loop:
-
Writes a file
-
Reads a file
-
Fails to confirm changes
-
Repeats indefinitely
-
-
Diff application frequently fails, usually because the AI can’t find the file at the expected location
Overall impression so far
To be honest, while I’m impressed by the ambition and potential of the AI tools in Wappler, I’ve struggled to find much practical, reliable use so far — particularly for Server Actions.
Most errors ultimately boil down to:
-
File not found
-
File not accessible
-
Schema mismatch
-
Incorrect or unclear file focus
My question
Am I doing something fundamentally wrong in how I’m using the AI Builder?
Or is the AI Builder — specifically around Server Actions — still quite early / unstable at this stage?
Any clarification on:
-
Current limitations
-
Expected workflows
-
Best practices (especially with Node.js + Docker)
would be greatly appreciated.
One additional observation: GPT-5.1 appears to work reasonably well when used in AI Agent mode, but not in AI Manager mode.
In contrast, GPT-5.2 does not seem to work at all in my setup, regardless of mode.
Is the AI Builder, in its current form, considered fully production-ready, or should it be treated as an experimental/early-access feature—particularly when working with Server Actions?
