How Antigravity became Cursor 2.0 (and not in a good way)

I think it’s no secret to anyone that when Antigravity first launched, it felt like a breath of fresh air for all of us—especially after dealing with the overly corporate and restrictive environment of IDEs like Cursor.

When I purchased the $20 Pro subscription, I was genuinely impressed by what Google had built. It honestly felt like a “Robin Hood” move in the AI space—a platform that actually valued solo developers and provided a reliable tool for deep, uninterrupted work.

Unfortunately, recent policy changes and the overall treatment of the user base have left me deeply disappointed. To give the Antigravity team some context, here is how the experience completely fell apart for me over the last few weeks:

1. The Pre-Conference 429 Nightmare For about 2–3 weeks leading up to the May 19th update, many users (including myself) were hit with constant HTTP 429 “Resource Exhausted” errors. This completely paralyzed my workflow. Even though my dashboard showed plenty of available quotas, I couldn’t use a single AI model. We effectively paid for a subscription we couldn’t use, with zero communication.

2. The May 19th Overhaul: The Illusion of Progress The recent conference promised us a “door to a new world,” but the reality is much bleaker. Replacing Gemini 3.0 Flash with version 3.5 felt like a massive step backward in reasoning capability.

3. Shifting Infrastructure Costs to the Users The shift from a transparent request-based limit to a hidden “compute-used” tracking system completely ruined the user experience.

Based on Google’s initial marketing, it feels like we, the lower-tier $20 users, are supposed to just be grateful that we are even allowed to touch this IDE. But how can we call this a “miracle” when Gemini 3.5 Flash has turned into a hallucination machine? Instead of writing clean code, the model now generates massive walls of text (often 10+ paragraphs explaining obvious things).

Because of this extreme verbosity, combined with the invisible background context scanning and micro-step agent queries that Antigravity runs behind our backs, the Pro quota evaporates faster than it ever did when using the heavier Gemini 3.1 Pro model.

Conclusion: Let’s look at the raw hypocrisy of this update. With this new “compute-used” system, Google legally obliged us to cover their rising hardware and infrastructure costs out of our own pockets. Yet, when the tables were turned, nobody covered our financial or productivity losses when we were completely locked out of the IDE due to their constant, endless server errors—like the infamous HTTP 429 nightmare that paralyzed us for weeks.

We came to Antigravity to escape corporate greed. Instead, we got a downgraded model that eats up its own quota by being overly talkative, and a hard lockout that forces us to either sit in a 5-day ban or upgrade to a $100–$200 Ultra plan just to keep working. If the goal was to push independent developers back to Cursor or Claude Code, congratulations—you are succeeding.

I really want to believe that Google will fix this situation. I understand the obvious fact that there is no such thing as a free lunch, but you were the ones who promised us this value in the first place. I believe that the more threads like this we create, the higher the chance they will actually fix it. I generally love Google and I am grateful for their products. I am the kind of person who never posts on forums or contacts support. But this absolute nightmare forced even someone like me to write a post, just to get a concrete explanation of what exactly I am paying my hard-earned money for.

And before anyone tries to justify this with the usual corporate line that “the paid version offers advanced features that explain the performance difference” — let’s be real for a second.

What features are we talking about? If you look at the other threads right now, everyone is complaining about the exact same things. The automated background agents are completely unpredictable—they loop endlessly, scan the same files over and over, and wipe out a user’s entire weekly token quota in 20 minutes for a simple task. That’s not an “advanced workflow,” that’s just a broken system architecture.

Half the forum is currently drowning in bug reports because the recent update broke workspaces, messed up local settings, and threw constant connection errors. Even the creative creators in adjacent topics are getting hit—their harmless prompts get flagged by over-sensitive automated filters, nothing gets generated, but their credits are still deducted without a refund. That is simply unacceptable for a paid product.

The bottom line is simple: if a completely free, standard web-chat is more reliable, predictable, and useful for daily work than a specialized environment people are expected to pay for, something is fundamentally wrong. Google needs to stop using paying developers as beta-testers, fix the opaque token drain, and give us back a functional tool.