As a long-term Google power user and a Ultra subscriber, I am writing this to express my absolute frustration with the persistent service instability.
To be blunt: Antigravity has become rather than a productivity tool.
Over the past few days, the experience has been unacceptable. I am raising the following critical issues for the product team to address:
1. The “Model Capacity” Trap for Paid Users I am paying for the highest-tier Ultra subscription precisely for prioritized access. Yet, I am repeatedly greeted with MODEL_CAPACITY_EXHAUSTED errors. If paying customers are forced to compete for basic compute resources—and are frequently throttled or locked out during non-peak hours—what exactly am I paying for?
2. Silent Failures and Infinite “Running” States Countless Agent tasks get stuck in a Running loop for 20+ minutes without logs, output, or timeouts. In a fast-paced environment involving complex code audits and refactoring, this level of unpredictability is a massive drain on resources. We are here to ship code, not to act as unpaid Beta testers for Google’s infrastructure.
3. Subpar Infrastructure Compared to competitors like Claude Code and OpenAI Codex, Antigravity’s connection stability is disgraceful. Frequent 503 Service Unavailable errors, authentication conflicts, and random disconnects have become the “new normal.” For a company that prides itself on global-scale infrastructure, this is an embarrassment.
4. Broken Multi-Account Authentication The bug where the system fails to recognize a Pro/Ultra license when multiple Google accounts are logged into the browser is still rampant. This is a fundamental UI/UX failure that should have been patched months ago.
My Questions to the Google Antigravity Team: Do you genuinely value the developer community? If Google cannot provide a reliable, deterministic environment for production-level development, we have no choice but to migrate our entire workflow to the Anthropic or OpenAI ecosystems.
Developers pay for efficiency and certainty, not for a “half-baked” experiment that quits when we need it most. I expect a clear response: What is the roadmap for fixing these server-side fluctuations, and how will Google compensate users for the significant downtime these days?