Recently, Google AI Studio’s Build mode seems to be behaving incorrectly.
Since today, when working with larger existing code files and requesting only non-functional changes (for example: “optimize”, “refactor for readability”, or “format the code without changing functionality”), Build mode no longer preserves the original functionality.
Example:
-
Original file size: about 1600 lines
-
Request:
optimize / clean up / refactor style without changing behavior
-
Actual result:
-
Output is much shorter (around 400–500 lines)
-
There is no warning about truncation or omission
-
Instead, the model silently removes large portions of existing logic and features
-
It still reports the task as “completed”
-
So this is not a token-limit truncation issue.
It appears that Build mode is:
-
Making its own decisions about which parts of the code are “unnecessary”
-
Performing implicit feature removal
-
Producing output that is not functionally equivalent to the original code
This is especially problematic for:
-
Refactoring large frontend files (React / TSX / Vue)
-
Formatting or restructuring existing project files
-
Making small edits to production code
Combined with other recent issues (Reload button missing, “data was moved” errors, preview desync), this suggests a regression in Build mode’s generation or state management behavior.
Questions for others:
-
Has anyone else noticed Build mode removing functionality during refactors?
-
Was there a recent change to its output or simplification strategy?
-
Is this a known issue?
At the moment, Build mode feels unsafe for large files, because it no longer guarantees functional equivalence when asked to perform non-functional edits.