I’ve been using Google Antigravity heavily lately, and while it’s generally a fantastic IDE, I’ve run into a severe issue with how the agent handles server traffic spikes.
As you can see in the attached image (Screenshot 2026-06-20 at 16.39.27.png), I had a task running on Gemini 3.1 Pro for 2 hours and 9 minutes. Right in the middle of it, I hit the following error: “Our servers are experiencing high traffic right now, please try again in a minute.”
The biggest issue isn’t the traffic itself; it’s how the IDE handles the failure:
-
Massive Token Waste: Because the agent hard-terminated, the feature was never finished, and the model couldn’t figure out how to resume from the broken state. This single failed run burned through nearly 20% of my weekly tokens in less than 12 hours with absolutely zero usable output.
-
Project Corruption: The termination left my codebase completely broken. Instead of gracefully stopping, it abandoned the project mid-edit, forcing me to pull a working version from GitHub and lose all the progress.
Feature Request / Recommendation: When the servers experience high traffic, the job should absolutely not be terminated, especially on long-running tasks. Instead, please consider implementing one of the following:
-
A Queue/Pause System: If the server is busy, put the active agent into a paused queue and automatically resume it when traffic drops, rather than hard-crashing.
-
Auto-Stashing/Rollback: If an agent is going to terminate due to an external server error, it needs to automatically stash its incomplete changes or roll back to the last stable state so it doesn’t destroy the project files.
Wasting a huge chunk of a token quota on a server-side failure that also breaks the local project is incredibly punishing for developers. I hope the team can look into better error handling for this!
Thanks.
