I’m a Google AI Pro subscriber and a cloud architect who regularly uses multiple AI coding agents, including Claude Code, Codex, and Gemini.
I’ve been testing Gemini 4 Argon, and unfortunately, I’ve encountered three significant usability issues that make it difficult to rely on for real-world development workflows.
- Gemini 4 Argon fails to recognize uploaded audio files in the web app
When I upload an audio file to the Gemini web interface, Argon fails to recognize or process its contents.
This is especially frustrating because audio input is an important part of a multimodal AI workflow.
Expected: Argon should recognize and analyze supported audio files uploaded through the web interface.
- Repeated approval prompts during GitHub pull request workflows
When using Argon in an agentic development workflow to create and submit GitHub pull requests, I repeatedly have to approve actions.
The excessive approval prompts interrupt the workflow, reduce productivity, and seriously degrade the developer experience.
I’m not asking for security protections to be removed. I’m asking for a better balance between appropriate security controls and efficient task execution.
Expected: Safe, routine actions should be handled according to the user’s configured approval policy, without unnecessary repeated confirmation.
- Usage limits are exhausted far too quickly
Argon’s available usage quota runs out very quickly during normal development tasks.
As a paying Google AI Pro subscriber, this makes the model difficult to use consistently for serious work.
Even when the model performs well, a tool that becomes unavailable in the middle of a task is not reliable enough for production-oriented development workflows.
Expected: More practical usage limits, clearer quota visibility, and predictable resource consumption.
Conclusion
Gemini 4 Argon may have strong underlying model capabilities, but model intelligence alone is not enough.
For developers, the end-to-end experience matters just as much: multimodal reliability, agent autonomy, workflow continuity, and usable quotas.
I hope the Gemini and Antigravity engineering teams can investigate these issues and prioritize improvements.
I’d also be interested to hear whether other developers are experiencing similar problems.