Google’s Gemini AI Is Making My Smart Home Less Useful

Last night my wife asked Gemini to turn off the reptarium lights in our living room. She asked twice then instructed me to try since it listens to me better. I did the same and after a few tries it finally listened.

That should be one of the lowest-stakes requests a smart home can handle. No nuance. No research. No reasoning chain. Just: lights off.

Instead, I got a message about getting medical help and that Gemini can't be reliably trusted for this. I tried to get a screenshot of this but my phone it dismissed before I did.

I wish that felt like an outlier, but lately my experience with Google devices has become a steady reminder that “smart” often means unpredictable in the most mundane moments. Voice assistants have always had rough edges, but this feels different. Google keeps replacing simple, working behavior with something more confused, more evasive, and somehow less useful.

The door locks are the best example.

I have 3 Nest x Yale locks. These are not some obscure third-party devices duct-taped into a smart home setup. They are Google’s own ecosystem. For years, I have been able to ask whether the doors are locked, or ask Google to lock them. That is exactly the kind of thing a smart assistant should be good at.

Now, depending on the day, Gemini will tell me it cannot tell whether the doors are locked. Or it will say it cannot lock doors. Or it will behave as if this is a new and impossible request, despite the fact that it has been doing exactly that for years.

That is what makes the experience so frustrating. It is not that I am asking for some futuristic, speculative feature. I am asking for basic continuity. The devices already exist. The integrations already worked. The commands were already part of daily life. And now the system feels like it is forgetting how to operate the house it was supposedly built to manage.

A smart home assistant has one job before all others: reduce friction. If I ask it to turn off the lights, set a timer, play music, stop an alarm, check a lock, or lock a door, I do not want a personality. I do not want a safety lecture triggered by a hallucinated interpretation. I do not want to debug intent classification while standing in a hallway. I want the thing to do the thing.

Google’s problem here feels bigger than one bad response. The company keeps pushing AI deeper into products where trust is earned through boring reliability. Gemini may be impressive in a demo, but inside the home, impressive is less important than dependable. A lock that works every time is better than an assistant that can summarize the internet but suddenly cannot tell me if my own front door is secure.

That mismatch is what makes the experience so maddening. These devices live in intimate spaces. Bedrooms. Kitchens. Kids’ rooms. Front doors. Late nights. Early mornings. The tolerance for nonsense is much lower there than it is in a browser tab. When a smart speaker gives an absurd answer, it is not just a funny AI fail. It breaks the illusion that the home is under your control.

And with locks, it goes beyond annoyance. If the lights do not turn off, that is irritating. If the assistant cannot tell me whether the doors are locked, that changes how safe the house feels. It turns a convenience into a question mark.

I do not need my assistant to be brilliant. I need it to be competent. I need Google to remember that home automation is infrastructure, not theater.

Until then, every simple command comes with a tiny gamble: will the lights turn off, will the doors lock, or will my house decide this is a healthcare conversation?

If things don't improve by the end of this year thousands of dollars of Google Nest equipment will be replaced and I will stop buying Pixel devices. I have had enough.