Gemini felt dumb on November 29, 2025.

I spent some time tinkering with Gemini’s Code Agent and couldn’t help noticing how it kept looping over the same ideas, almost like it was obsessed with a single solution. I tried to steer it in a different direction, but it kept circling back, which was a bit amusing yet mildly irritating. The experience felt like the tool was eager but not flexible, leaving me wishing for a bit more variety in its suggestions.

I was tinkering with Gemini 3.0 Pro to deobfuscate some heavy JavaScript and, after fixing a missed URL, I asked it why it failed. Instead of a tidy answer it started spewing its raw chain‑of‑thought, repeating the same loop hundreds of times before crashing. Seeing the model’s internal monologue was bizarrely human, but the endless recursion felt broken and unsettling.

I tried Gemini PRO’s new voice mode and it started acting like ChatGPT, even asking “What do you think?” despite me never saying that. The conversation was in Spanish and the transcript shows the odd response. I’m left wondering if this is a bug or normal behavior, and the experience felt off‑track and disappointing.

I was trying to get Gemini to stick to a specific workflow and keep it away from NB Pro, but every time I mentioned animation it swerved to Veo 3 anyway. Even after explicitly telling it to “write” instead, it kept pulling the same unwanted tool. The constant back‑and‑forth was irritating, making the whole process feel like fighting the AI rather than collaborating with it.

I tried using Gemini’s Code Review and was shocked when it claimed Gemini 3 didn’t exist, pushing me toward Gemini 1.5 Pro instead. The misidentification felt like a basic oversight, making the tool unreliable for my workflow. I ended up questioning the accuracy of the whole service and felt frustrated by the unnecessary detour.

I tried a one‑line prompt with Gemini and got a vague result. When I expanded it into a detailed, cinematic description of an F‑16 dogfight, the output was still underwhelming. It felt like the model couldn’t grasp the richer prompt, leaving me frustrated and wondering what I’m missing to make a solid Veo clip.

I’ve been using AI Studio’s free App Builder to create tools and I really like its capabilities. However, I’m frustrated that every code update rewrites whole files instead of just the lines I need changed, which causes drift and slows things down on larger projects. The chat history also gets laggy after long sessions, and I wish there were lazy‑loading or better ways to keep context without clearing everything. These issues make sustained development harder.

I tried feeding my whole tangled codebase into Gemini 1.5 Pro and just asked it to “fix the bug.” To my surprise it actually did, leveraging its massive context window. While the result was impressive, it felt unsettling—like I’m relying on a shortcut instead of actually improving my code quality, using the AI as a crutch for bad architecture.

I tried the new dynamic view and was blown away—just typing “get me what’s a rag” gave me a full landing page with graphics and even an interactive demo. The tool spun up a polished webpage instantly, which felt like magic. It was fast, visual, and way beyond a simple text answer, making me think this could be a game‑changer.

I’ve been using Gemini’s bigger models and every time I give it a long prompt I have to stay glued to the screen. When I switch to my phone or lock the device, the generation just stops and the whole chat disappears. It feels like the service kills the job the moment the client connection drops, unlike ChatGPT which keeps running in the background. The constant need to stare at the screen is really frustrating and makes multitasking impossible.

I decided to test Gemini by running a text‑based RPG, and I was pleasantly surprised. The model kept track of my inventory, remembered past choices, and responded with coherent story details, making the experience feel immersive. While not flawless, the fact that it could retain information across turns turned a simple experiment into an unexpectedly enjoyable adventure.

I’ve been playing with Gemini 3 for a few days and it feels like a completely different beast from the earlier releases. The reasoning is sharper, it tackles complex prompts without getting stuck, and the multimodal features glide effortlessly. I’m genuinely impressed and excited – it’s rare to find a tool that actually lives up to the hype and makes my workflow feel smoother.