This above is the latest bug on my Windows machine :)
Mac doesn't have it. It has the disco bug.
Did your Google Antigravity also do some disco today?
I was deep into a Fast-API refactor for one of my extension, trying to navigate a tricky bit of Python logic.
Yesterday, when I hit the left arrow key to move my cursor, suddenly my IDE violently switched from dark mode to light mode.
For about ten seconds, my M3 screen looked like a strobe light at a club.
And today, it's the ide jumping across the screen on Windows.
I just read.
It turns out I'm not the only one throwing an impromptu rave at my desk.
The community confirms that the latest update has introduced a suite of baffling bugs that are severely disrupting developer workflows.
The most common issue being reported is the one I experienced: the theme-switching bug.
For some users, simply pressing the left and right directional keys triggers the IDE theme to flip between its light and dark states.
If you are quickly navigating through a line of code or trying to highlight a specific block, your screen literally flashes, turning your workspace into a disco.
It's a UI glitch that makes the editor actively painful to look at during standard navigation. Please don't come at me and say
Shashwat, aren't you used to these bugs by now?
While the flashing lights are annoying, the actual workflow killer is the silent memory wipe. Several developers noted that their conversation history for the last 2 weeks has completely vanished.
One user, on the Google community explicitly stated they use the tool daily, yet all their recent chat context and different workspaces in agent manager mode were erased upon launching the IDE. For those of us who rely on Antigravity to maintain deep context on complex repositories, losing a week and a half of architectural discussions and debugging steps is brutal
Luckily, it hasn't happened to me.
The missing history feeds directly into the third major issue:
the context crash loop
According to community reports, the agent is currently struggling with long context windows.
After a lengthy conversation, the agent will simply error out and drop all memory of the current task. In a stable build, the standard developer response would be to restart the IDE and reload the chat history.
However, because the update is actively failing to save conversation history, logging out to fix the crashed agent means you lose that active context forever.
You are essentially forced to start from scratch every single time the underlying model throws an error.
Between the UI strobe lights, the missing chats, and the fatal crashes, the community sentiment is incredibly frustrated.
One user even commented, "I'm pretty sure Google created this and doesn't use it" on Reddit.
After the whole ban-wave remediation drama last month where developers (including my friend Sumeet) had their accounts nuked for trying to optimize their workflows, you'd think the platform would focus entirely on raw stability.
Until Google pushes a hotfix for this, I'm keeping my backend work safely housed in my local setup.
I'll rely on my local terminal configurations and standard Claude Code to handle the heavy lifting. We need reliable tools, not a disco ball in our development environment.
I swear tracking these updates is a job in itself, lately.
Here's the list which I've built and keep adding on.
In case we are meeting for the first time, come over here, it'll be worth the roller coaster of articles that are gonna come up in the next few weeks.