Gemini felt dumb on January 18, 2026.

I tried using Gemini for a coding task and it was a nightmare. It never asked clarification questions, ignored my .agent files, and even when I gave explicit instructions it “improved” them into worse results. It kept getting stuck, then produced something completely different. In contrast, Claude admitted its limits and helped me finish the project, making the whole experience far less frustrating.

I was working on a massive 750k‑token conversation in AI Studio and everything was fine, but now each time I refresh the page the latest prompts and answers vanish, reverting to the start. It’s infuriating because I lose all the recent context I’d built, and I can’t figure out a fix or if anyone else has this problem.

I keep hitting a CENSORED error with Nanobanana Pro on almost every prompt, even for something as simple as a back‑facing woman in a bikini. The interface looks great, but the over‑aggressive filtering is driving me nuts. I can’t generate SFW images I know the model can produce elsewhere, and I’m stuck looking for a less‑censored platform without running anything locally.

I noticed that sometimes the model skips its usual “thinking” step and just gives me the answer right away. After being offline for a few days, I was surprised to see this shortcut in action. It felt oddly fast, but I wasn’t sure if it was a glitch or an intentional tweak, leaving me a bit puzzled about the tool’s consistency.

I tried Gemini’s voice transcription and it was a nightmare. Every time I spoke, it mangled my words, cut me off mid‑sentence, and sent half‑baked queries to the model before I was finished. I felt frustrated watching it jump the gun, especially compared to ChatGPT’s Whisper, which lets me pause, think, and transcribe perfectly. The whole experience made me doubt Google’s claim of having the best free voice tech.

I asked Gemini to make an image, got a cool picture, then just replied with “cool” or “nice”. Instead of a normal chat reply, it automatically spawns another image—usually the exact same one pixel‑for‑pixel. It happens about ninety percent of the time, which is extremely irritating. I’m looking for a fix so it stops auto‑generating duplicate images.

I set up a prompt telling Gemini never to say “Nano banana,” hoping it would respect my request. Instead, it kept slipping that phrase in, which was really annoying. I felt the model wasn’t listening to my instructions, turning a simple tweak into a frustrating back‑and‑forth.

I tried to use Gemini’s Personal Intelligence to pull a whole folder from Google Drive so I could chat about its files. Every time I gave it clear instructions, it would stall, spin up new search strategies, and mostly hallucinate or return nothing. Only about one in ten tries actually fetched the content, and even then it took a minute. The whole process was slow and frustrating.

I asked Gemini where the CapCut project folder is on Windows, and instead it spiraled into a nonstop monologue about shame, looping the same phrases for minutes. The endless, repetitive output left me feeling frustrated and stuck, wasting time and giving me no useful answer. It was a painful, unhelpful experience.