CRITICAL WARNING TO ALL DEVELOPERS: It is highly advised NOT to use this tool in production or sensitive projects right now. The severe risks and destructive blind actions executed by this broken update pose a massive danger to any real-world project. Please avoid using the application in its current state to protect your codebase.
Hello,
Why do we need “vibe coding” when we already have the ability to complete our workflows conventionally? The answer is simple: to focus more on creativity, save time, and increase productivity to build actual problem-solving projects—that is the core purpose of an AI tool. If the tool destroys the work we spent time on, or operates counterproductively to the very benefits it is supposed to provide, it becomes completely illogical and entirely unreliable.
What happened in this Antigravity update is utter chaos that you must fix immediately and urgently. If Google engineers operate with this level of chaos, there is no point in releasing products that offer no real value and only serve to waste the time and money of paid subscribers.
I am facing severe, workflow-breaking issues since the latest update. This update feels catastrophic and has completely ruined the IDE experience. Here are the main issues:
Reversed Quota Consumption: The quota system is broken. My Weekly Limit is being consumed rapidly, while my Five Hour Limit remains completely untouched at 100%. The system is draining the weekly limits instead of the 5-hour quota.
Unresponsive UI & Infinite Loops: The user interface is practically frozen. When trying to open certain projects, the sidebar gets stuck in an infinite loading loop displaying “Checking your setup…”, preventing access to the project entirely.
Forced Auto-Update on Downgrade: I attempted a manual downgrade to a previous stable version, but the IDE forcefully auto-updates immediately upon launch, making it impossible to bypass the current broken version.
Opaque AI Actions & Missing Diff Viewer: There is no longer any system to track what the AI modifies in the code. The activity log shows the AI actively working in the background (analyzing, creating, and editing files), but clicking on these log entries does absolutely nothing. There is no Diff Viewer to see what was actually changed. We are only given a strict, binary “accept” or “reject” option. Expecting developers to blindly accept unseen code changes destroys work and introduces critical risks.
Broken Agent-Project Connection (Missing “Problems” Integration): The connection between the AI agent and the project workspace is completely lost. Previously, I could go to the “Problems” tab in the terminal pane and pipe all errors directly to the agent for analysis and troubleshooting. Now, I cannot do this at all, making the AI useless for debugging.
Removed Update Console: I can no longer manually check for updates or view the update environment console to see what is happening under the hood. This accessibility has been completely removed.
Chaotic UI/UX Design: Cramming all functionalities and tools into sidebars has resulted in a cluttered, chaotic design and a terrible user experience.
Removed Voice-to-Text Feature: The integrated voice typing/dictation feature has inexplicably disappeared without any logical reason, removing a vital productivity and accessibility tool.
At this point, the program just feels like a standard VS Code, and frankly, I would be much better off using normal VS Code without any AI services rather than dealing with this chaotic environment.
A final note to the development team: I always understood that updating an application means improving existing features or adding new ones while maintaining a stable core interface. It is completely unreasonable that every update drastically changes the application itself and breaks what used to work smoothly.
Are there any immediate workarounds to stop the auto-update, fix the quota consumption, or restore the Diff viewer?
Thank you for sharing this detailed feedback on the Antigravity IDE (v2.6.0) update. We have noted your feedback regarding voice input and the updated IDE interface and shared it with our team.
For the three immediate workarounds you asked about:
Stop auto-updates: In Settings (Ctrl/Cmd+,), set Update: Mode (update.mode) to manual or none.
Quota consumption: Check Antigravity Settings → Account or run Antigravity: Reset Onboarding from the Command Palette (Ctrl/Cmd+Shift+P) and sign back in if your Google AI Pro/Ultra session was reset during the update.
Diff viewer & setup hang: Run Developer: Reload Window from the Command Palette and ensure Toggle Inline Diff (antigravity.enableInlineDiff) is enabled under the Antigravity panel’s ... (More Actions) menu.
I am facing serious issues with AI agent code modifications after the latest update. The agent is making incorrect and unnecessary changes to existing code, often modifying working functionality instead of following the actual instructions.
The situation is becoming worse because the Accept/Reject options and proper Diff Viewer are missing. We cannot properly review what the agent has changed before those modifications affect our projects.
This is extremely frustrating. We need the ability to inspect every code modification, review individual changes, and accept or reject them before they are applied.
Please restore the previous Accept/Reject functionality and Diff Viewer immediately. Developers should never be forced to blindly trust AI-generated code, especially when the agent is not making the correct changes.
Antigravity is supposed to improve productivity, not create more work by breaking existing code. The current experience is becoming worse with every update.
Actually you are entirely correct the previous version is far more better than the new Version, its literally fully broken.
The files the Agent Edits does not Appear in Review Changes. if it edits 8 files then only 3-4 Appear in Review Changes.
Okay another issue is when sending a message it disappears without a reason
I type that message then click enter and it’s queued there and a send all button and there is nothing the agent is doing anything then I try a new chat and in the new chat sometimes I would want to stop the chat it just does not stop no matter how much I click stop send all queued messages nothing works
Also the diff workflow has a issue which ruined my edits that I have been working on for a whole day which is it duplicates the lines like it would change somethin and the old code is still there if antigraphty ide shuts down in that state or I edit manually any other lines chaos comes with overwrite message and two lines differ so now
I’m forced to just watch the agent complete everything with no tabs open and in the same time the only way to check changes is the github desktop.
Even if it was vibe coded the latest update it wouldn’t be this much chaos like how a whole company does not even have testers or the developers test after thier job
Also I notice soooo much lag with big chats which was already solved 3 months ago.
When I use chatgpt desktop or claude code I just wonder why do
I’m even using it, and another thing is it’s going back years behind it’s now even dummer and stupider than the old copliot that was in vscode.
Also it has excessive searching it searchs for 12 minutes or one time a whole hour just to change a simple theme thing.
And then it just does not even do the job after all that searching because the context then gets compacted and I go again
I was always a big fan and a user of antigraphity and would recommend it to everyone but now it’s the worse product ever
I hope u revert the old antigraphity instead of being lazy to manage two apps the antigraphity vscode extension or the ide.
Antigravity is suffering from severe context confusion. I’ve created a dedicated project repository and placed all relevant chats into a single project folder-yet, it somehow ends up reading an Agent.md file from an entirely different project and mistakenly writes data into that one instead. Furthermore, the interface on my machine is mixing up tasks; even though I have two separate chat windows open, clicking on one reveals that it’s actually executing a task belonging to the other-how could such a major bug even happen? To make matters worse, my weekly usage quota is depleting rapidly, while the 5-hour limit remains completely untouched. I’m finding it incredibly frustrating to have to spend so much extra time constantly correcting Antigravity’s work, bit by bit.
Currently, Antigravity IDE doesn’t seem to have proper options for reviewing code changes. Previously, we had features to navigate through modified files, accept or reject individual changes, and use a proper diff viewer. However, these features seem to be missing now.
Also, the Gemini 3.8 Flash model feels slower than previous Flash models and requests significantly more permissions compared to other models.
