I Paid Google for a Pro Plan and Got a Fancy Spinner

I want to start by saying I gave this a real shot. Not a weekend experiment, not a frustrated rage-quit after one bad session. I paid for the Pro plan, I adjusted my workflows, I swapped models, I read the forums, I did everything short of sacrificing a keyboard to make Google Antigravity work for me. And I'm writing this because I think a lot of developers are quietly in the same place but haven't said it out loud yet.

The rate limits are the obvious complaint and everyone knows about them by now. But what nobody talks about enough is how deeply insulting the structure is once you understand it. Google doesn't just give you a daily limit. There's a 5-hour rolling cycle sitting on top of a 7-day hard cap that they didn't even document for months after they introduced it. So you start a session, feel like you're flying, hit the 5-hour reset, feel okay about it, and then somewhere around day two you slam into a wall that won't move for a week. I burned through my weekly quota in a single focused afternoon of feature work. One afternoon. On a plan I'm paying twenty dollars a month for. The app's response was to show me an "Upgrade to Ultra" prompt. Two hundred and fifty dollars a month. That's not a product improvement, that's a ransom note.

But here's the thing that actually broke me, and it wasn't the rate limits at all. It was Gemini itself. The model that's supposed to be powering all of this has an 88% hallucination rate when it encounters something it doesn't know. Not 88% hallucination overall, specifically 88% of the time it gets something wrong, it doesn't say "I'm not sure" -- it says the wrong thing with complete authority. I watched it delete an entire function and replace it with the same broken line seven times in a row. Not a variation, the exact same wrong line. And between each attempt it told me the issue was resolved. I thought I was losing my mind. Turns out I wasn't, this is documented behavior and there's a whole thread on the developer forum about it.

What makes it sting more is that Gemini is genuinely good at one thing: writing clean, polished English. Ask it to draft an email, summarize a document, explain a concept -- it's smooth, it's articulate, it sounds great. The moment you give it a real codebase and ask it to do something non-trivial, it hallucinates imports that don't exist, invents API methods, and confidently restructures things it was never asked to touch. I've started making a Git commit before every single prompt just as a survival mechanism. That should not be normal.

After months of this I switched to Claude Code. I'm not going to make this a long comparison because I don't think it needs one. Claude is the best product I have seen in this world after Apple. That's a strong thing to say and I mean it. It tells you when it's unsure. It doesn't rewrite things you didn't ask it to rewrite. It reads your project context from a file you write once and actually retains it through the session. It broke 80% on SWE-bench, which is the closest benchmark we have to real-world engineering work, and you can feel the difference the second you use it on something that actually matters. The Apple comparison isn't accidental -- both companies decided they'd rather do fewer things with real craft than ship everything and patch it later.

The reality is that Google’s product engine is starting to feel like it’s suffering from a fundamental lack of both skill and taste. Their products are consistently degrading day by day because they’ve lost the ability to distinguish between what’s worth shipping and what’s just noise.

Google built Antigravity to chase Cursor and Windsurf and ended up building something that's worse than both. They have the infrastructure, they have the talent, they have a model that leads almost every benchmark on paper. And somehow they shipped a product that makes paid users feel like beta testers on a degrading free tier. That's not a technical failure. That's a product philosophy failure. And no amount of weekly quota resets is going to fix it.