Gemini felt genius on November 13, 2025.

I managed to recreate Balatro on mobile using Gemini 3 in just two prompts, adding mult and points scoring, jokers, and progressive difficulty. However, I hit a snag when the share link clipped the bottom of the screen, so I had to spend three more prompts tweaking the UI. It worked, but the extra hassle was a bit frustrating.

I was trying to fix a tiny button issue in my web app using Google AI Studio, but instead of just tweaking the code it completely broke the entire project. Every time I retried, the AI rewrote large sections, forcing me to constantly restore backups. The tool's behavior felt risky and extremely frustrating, turning a simple debug into a nightmare.

I tried Gemini 3 by building a simple boxing game in a single HTML file, just to see how well it could handle the task. The AI put together the code quickly and it actually worked, which was surprisingly smooth. I felt impressed that it managed the whole thing so cleanly, making the experiment feel successful and enjoyable.

I tried Gemini 3 and was immediately struck by how much better the role‑play feels compared to version 2.5. The responses were smoother, more context‑aware, and the whole interaction felt a lot richer. It was a noticeable jump in quality – the upgrade felt huge and made my experience genuinely enjoyable.

I keep hitting 503 Service Unavailable or “model overloaded” messages every time I call Gemini via the genai Python library. It’s happening to many developers, and even after adding exponential‑backoff retries the problem persists. The constant failures are really frustrating and make the API feel unreliable.

I tried Gemini 3 on a Bloons Tower Defense mobile project and, after just three prompts, it nailed the upgrades and wave system. The results were flawless and blew my mind—so smooth and powerful that it felt almost magical. The tool’s performance was beyond what I expected, leaving me thrilled and amazed.

I was blown away when I tried Gemini 3.0 Pro in Canvas – with a single prompt it churned out an entire Android OS layout. The speed and completeness felt almost magical, like the tool just “got” my vision instantly. It turned a daunting task into a simple click, leaving me amazed at how powerful it truly is.

I tried using Gemini for a simple CSS tweak and it wiped out my whole code, leaving a placeholder line. It then botched an Excel duplicate‑removal request—first refusing the upload, then claiming it’d sent a file that never appeared, and finally giving me a broken HTML uploader. Even marketing image requests ended with irrelevant Shutterfly links. Compared to Claude, Gemini felt unreliable and lagging, turning a routine project into a frustrating saga.