Gemini felt dumb on October 29, 2025.

I used Gemini to build GDesk, a free Google Services launcher that runs in the browser with bright and dark modes and no login required. The AI helped me piece together the code and design, and the result stores wallpapers and notes in the cache. I’m thrilled with how smoothly it guided me, and sharing the repo and live demo feels like a win.

I’ve been using AI Studio and started noticing that it rarely “thinks” before it spits out an answer, even though I have the thinking option turned on. It feels like the responses are rushed and superficial, which is pretty frustrating. I’m wondering if paying for the service would fix this or if the issue is just how the tool works now.

I tried using nano banana, seedream 4, and GPT Image 1 to create a trolleybus with its pantograph lowered, but every attempt fell short. The AI kept producing generic buses or completely missed the lowered pantograph detail, leaving me stuck and annoyed. It felt like the tool just didn’t understand the specific request, turning what should've been a quick visual into a frustrating trial‑and‑error session.

I fed Gemini 2.5 Pro massive scripts—each about 150 k characters—expecting it to handle the load. Instead, it just replied, “I am going to cry.” The response felt like a shrug rather than a solution, leaving me annoyed and questioning whether the model can even process such huge inputs. The whole exchange was surprisingly underwhelming.

I switched to Gemini after canceling ChatGPT, hoping for a smooth upgrade, but most answers lagged and ultimately didn’t work. The delays were irritating, and the outright failures left me questioning whether to stick with Gemini Studio or go back to a ChatGPT subscription. I’m frustrated and uneasy about the future of the service.