I Fired Google

That landed because lots of people report the same breakage across Google Home, Android Auto, and phone assistant flows. The recurring complaint is not that the old assistants were brilliant. It is that they were predictable. Users could learn the quirks and phrase commands to get the result they wanted. With Gemini, the failures feel probabilistic and opaque. Asking for a song now plays something merely related. Calling a spouse may trigger follow-up questions, fail to find the contact, or call the wrong person. Replying to messages, muting navigation alerts, setting timers, and identifying music all got slower or less dependable. Several people said the same drift is visible outside voice assistants too, especially in captions and search-style responses that now prefer plausible language over exact matching. The thread keeps coming back to one product lesson. Voice assistants were only ever truly valuable for a short list of repetitive, hands-busy tasks. Music, alarms, timers, lights, navigation, reminders, and weather came up again and again. LLM upgrades attacked exactly those flows by adding chatter, follow-up prompts, and semantic guesswork where users wanted intent parsing. That is why the mood is less "AI is bad" than "you broke the appliance." Once a tool sits in a kitchen, car, or bedside, people expect it to behave like a microwave or alarm clock, not a conversational partner trying to be helpful. People were not especially enthusiastic about Alexa as the escape hatch. Many said Amazon has its own version of the same disease, especially unsolicited "by the way" prompts and ads on Echo Show devices. The more useful conclusion was that the market left room for narrower systems. A few commenters described either going more analog or building local replacements with open wake word detection, on-device speech-to-text, and a local model tied to a safe set of actions. That fits the broader sentiment around Google here too. This was taken as another sign that consumer trust is eroding beyond technical users and into mainstream audiences who just want formerly simple tools to keep working.