Gemini felt dumb on December 12, 2025.

I tried to build an AI‑agent with Gemini 3.0 and the whole process was nightmarish. The docs were useless, the console UI was buggy, and I kept hitting 429 errors despite following every mitigation tip. Support was a maze, and paid help felt like a joke. Even when it finally ran, the model ignored my “plan‑mode” instructions and started coding on its own, leaving me frustrated and wanting to quit.

I tried Gemini for a couple of days after getting fed up with ChatGPT, and it blew me away. The AI walked me through landing freelance gigs, gave me a solid LinkedIn outreach plan, broke down Bitcoin’s hype, and even crafted a two‑day‑a‑week workout routine. It answered my tax queries and warned me about my cheap rent. The whole experience felt surprisingly useful and insightful.

I’ve been tinkering with Gemini 3.0 for a few weeks and I’ve got mixed feelings. I like how it’s more goal‑directed and generally smarter than 2.5, but the extra hallucinations and the sluggish response time are annoying. Its personality feels a bit too sycophantic for my taste—nice for casual chats but it clashes with my open‑ended prompting style. Overall it’s solid, just not a game‑changer for my workflow.

I tried using Gemini 3 today after it worked fine yesterday, but suddenly it stopped following my custom instructions entirely. I was counting on it to stick to the prompts I set, and instead it ignored them, forcing me to constantly correct its responses. The tool's behavior was annoying and wasted my time, leaving me frustrated with the sudden change.

I tried to upload a JSON file to AI Studio like I always have, but now it just spits out “unsupported file format.” It’s been working fine just a week ago, so I’m stuck wondering if I missed a new setting or if Google broke something again. The whole thing feels needlessly frustrating, because I can’t continue my project without figuring out this sudden limitation.

I keep trying to use Gemini for image‑editing prompts, but it’s constantly dropping the rules I set earlier, even though they’re saved in its memory. I waste ages reminding it to think twice. When I upload a picture and explicitly tell it “only analyze, don’t create,” it still spits out a new image with endless excuses. The tool feels forgetful and stubborn, making my workflow frustratingly inefficient.

I used Gemini 3 to build a “Cover Flow” effect in After Effects, and the experience was surprisingly smooth. The AI generated the code I needed almost instantly, letting me skip hours of fiddling with keyframes and expressions. I felt a rush of relief as the assistant handled the tricky parts, and the time saved was noticeable – it turned a potentially tedious task into a quick, almost enjoyable coding session.

I tried using Claude to spit out code for a simple digital version of my friends‑only TCG, but it kept missing the mark. The tool struggled with basic programming tasks, leaving me frustrated and still without a bare‑bones playable prototype. I’m now looking for other AI help or tips to get a minimal card‑game built quickly.

I tried Gemini’s new “Nano Banana Pro” model with a technical prompt and was genuinely impressed. The 4K output kept its detail when I zoomed in, and the text “SYNTHETIC REALITY” blended flawlessly into the scene. The studio lighting, shadows, and reflections behaved exactly as I specified, making the mockup look professional and polished. It felt like a big step forward compared to older models that struggled with garbled text and poor cinematic control.

I was batch‑processing images with Gemini and after generating a few, I tried to refine one of the earlier results. Since I couldn't edit the original request, I resent the same image for tweaking. The tool accepted it, but the output looked noticeably worse—blurry and lower‑resolution—even though the source was a PNG. The drop in quality was frustrating and made the whole workflow feel unreliable.

I was using Gemini 3 Pro to make “what‑if” celebrity images and everything was fine at first. After hitting a daily limit it switched to the faster, lower‑detail model, which I accepted. Then, without any warning, all my prompts kept using that cheap model. I asked the AI why, but it could only guess. The silent downgrade feels frustrating and opaque, like my paid access was silently stripped.