I have been using the latest AntiGravity IDE update on Linux, and while the UI design is well-managed and visually clean, the update has introduced several severe user experience regressions:
Loss of Gemini “Thinking” Process: In previous versions, the IDE displayed a detailed, real-time breakdown of Gemini’s thinking process. Now, it only shows a static “working…” text on a black screen. This lacks visual feedback, leaving users completely in the dark about whether the network is lagging, the request was received, or the IDE has frozen.
Code Review Panel Disabled During Execution: Previously, we could expand a summary popup at the bottom to review a list of edited files while Gemini was still working. Now, this information is completely hidden until the entire task finishes, breaking a seamless workflow.
Overly Aggressive Filters (False Positives): The latest version frequently blocks completely legitimate, normal requests with the error: “This request was blocked by Gemini’s filters…” simple, benign words are triggering the safety filters unnecessarily.
This update has unfortunately created a very frustrating experience on Linux. Restoring the live thinking feedback and the ongoing code review panel is critical for usability.
Thanks for flagging these regressions on Linux. We want to get this in front of the engineering team, and having a few extra details will help us narrow down what’s going on.
Could you share:
Your exact Antigravity IDE version (you can check this in the app menu) and which Linux distro and display server (X11 or Wayland) you’re running.
Which model and thinking level (Low, Medium, or High) you have selected in the model selector.
Any reproduction logs or error output from the developer console/terminal when the black ‘working…’ screen or filter block happens, along with a couple of benign example prompts that triggered the ‘This request was blocked by Gemini’s filters…’ message.
In the meantime, here are a few temporary workarounds you can try:
For the static ‘working…’ screen, try switching the thinking level (Low, Medium, or High) or selecting a different model in the model selector under the prompt box to see if the live feedback behaves differently.
Since the filter blocks happen on the model backend (reinstalling or clearing local cache won’t change them), try rephrasing the prompt slightly or switching to another available model in the model selector when a benign request gets blocked.
While the mid-turn review popup is unavailable, you can check which files are being modified in real time using your IDE’s Source Control view or by running git status / git diff in a terminal.
Once you share your version and logs, we’ll be able to investigate these regressions further.