If you were coding in December 2025,
you remember this:
the Antigravity agent was called a Senior Engineer.
It planned, it executed, it refactored.
Today, it’s an intern having a panic attack.
Note: Before I begin, I’d want to point out I’m on a Pro account, and not Ultra.
The Chain of Thought Loop
I gave the agent a simple task. Instead of writing code, it started talking to itself.
It repeated this 50 times before crashing.
What is this?
A classic “Context Collapse.”
To save money, Google seems to have reduced the inference compute for the Pro tier.
The model is trying to “think” (using Chain of Thought), but it doesn’t have enough compute to resolve the thought, so it resets and tries again.
It is an AI engaging in obsessive-compulsive behavior because it has been lobotomized.
The Silent Crash
When it’s not looping, it’s just giving up.
No stack trace. No explanation. Just “Agent execution terminated.”
I tried to reload. I tried to restart the environment.
But damn, I got the “Spinning Wheel of Death” for 10 minutes.
That’s it, I believe.
Antigravity is super unstable now.
You cannot build software if your IDE has a mental breakdown every time you ask for a refactor.
I might be done debugging the debugger. Or just use this for basic side projects.
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.