Critical Feedback on Google Antigravity (Gemini Pro 3.1 FAST) in Real Development Use

Google Antigravity
1

Dear Google Antigravity Team,

I am writing to provide detailed feedback regarding serious issues encountered while using Google Antigravity (with Gemini Pro 3.1 FAST) in a real-world development environment. The current behavior is not only unhelpful but actively disruptive to the development process.

1. Failure to understand project context and language

I was working on a Flutter/Dart project with 189 existing issues detected by the analyzer. I explicitly instructed the AI to fix all Dart-related errors. However, instead of addressing the Dart issues, the system:

  • Generated and executed unrelated Python test code

  • Completely deviated from the original task

This indicates a fundamental issue with context awareness and task alignment.

2. Destructive modifications to the codebase

During execution, the AI:

  • Deleted significant portions of existing Dart code

  • Introduced irrelevant and incorrect logic

As a result, the total number of errors increased from 189 to 891, including numerous repeated errors such as:

The function ‘AppSettings’ isn’t defined.
Try importing the library that defines ‘AppSettings’, correcting the name to the name of an existing function, or defining a function named ‘AppSettings’.

This is not error fixing — it is error amplification.

3. Clear performance gap compared to Codex

Under the same conditions, ChatGPT Codex:

  • Quickly identifies and fills in missing code

  • Fixes issues in a targeted and controlled manner

  • Preserves the integrity of the existing codebase

The difference in reliability and usability is significant.

4. Account risk when using Claude model

Another concerning issue:

  • The platform provides Claude as an optional model

  • However, using it may result in the associated Google account being flagged or even suspended

This presents a clear contradiction. If third-party models are officially supported, their use should not expose users to account risk. Otherwise, this creates confusion and undermines trust in the platform.

Conclusion

At present, Antigravity shows critical shortcomings in:

  • Context understanding

  • Task execution accuracy

  • Code modification safety

  • Third-party model reliability and account safety

I strongly urge your team to address these issues, particularly in real development scenarios. In its current state, the platform is not reliable enough to be used as a development tool.

I would appreciate clarification on whether these behaviors are known issues and if there are any planned fixes.

Sincerely,

Lei

1 Like
2

Hello @lei_mix ,

Welcome to AI Forum!

Thank you for bringing these concerns to our attention. Please be assured that I have shared your feedback with our internal team for further review.
We appreciate your continued patience as we work to enhance the Antigravity experience.