Affected version: Antigravity v1.22.2 (latest)
Account type: Google AI Ultra for Business (Workspace)
Models affected: All models tested, including Claude Opus and Sonnet
Description
After the v1.22.2 update partially addressed the previous “Agent terminated due to an error” regression, a new (or continuing) termination pattern has emerged.
Frequency: The agent terminates approximately 3-4 times per conversation before successfully completing its task.
Error shown:
“Our servers are experiencing high traffic right now, please try again in a minute.”
Key observation that suggests this is a bug, not real traffic:
When you click Retry, the agent resumes immediately - within 1-2 seconds. Real server congestion would not self-resolve in under 2 seconds. This behavior is consistent with a client-side error being incorrectly surfaced as a server traffic message.
Comparison to previous bug
| Period | Behavior | Usability |
|---|---|---|
| Before v1.22.2 | Terminated every few seconds | Unusable |
| After v1.22.2 (current) | Terminates 3-4x per session, retry works instantly | Barely usable |
The update reduced severity but did not fix the root cause.
Impact
- Paid enterprise users are experiencing repeated workflow interruptions
-
- Each termination breaks agent context, requires manual retry, and disrupts long agentic runs
-
- If reliability does not improve, users will migrate to alternatives (Codex, Claude Code) that do not have this issue
-
- The fake “high traffic” message is misleading and makes it harder for users to diagnose the real problem
Request to the team
- Investigate why the agent terminates mid-task and surfaces a “high traffic” error when retry succeeds instantly
-
- Fix the root cause - not just the error message
-
- Provide a status update on the ongoing termination regression that has now persisted for 5+ weeks
Related threads:
- PSA: “Agent terminated” bug is FIXED in v1.22.2 - was an IDE bug, not a model issue (original report - update: only partially fixed)