Apple Fumbled HomeView

Apple finally announced its smart display yesterday. It’s called HomeView, it costs US$299, and it’s a 7-inch square screen on a hinge with a speaker underneath. It knows up to six people by their face, shows your cameras and your calendar, and has a new Safari mode that turns recipe pages into steps with timers. If that sounds like an Echo Show, it’s because it is one. Amazon shipped the first Echo Show in 2017.

Apple had a shot at making the first AI device normal people want in their kitchen, and it’s blown it.

The problems start with the model. In January, Apple and Google announced that Apple’s next-generation foundation models would be based on Gemini, and Apple said that after “careful evaluation” Google’s tech was the most capable foundation it could find. I’d love to see that evaluation, because Gemini isn’t a frontier model in any way I care about.

Google’s newest one, Gemini 4 Argon, came out on 30 September and went generally available in the API yesterday, the same day as HomeView. Google’s launch table had it winning 13 of 19 benchmarks, which is great if your job is winning benchmarks. The rows it loses are the coding ones. It’s nine points behind Claude Opus 5.5 on Terminal-Bench 4.0 and more than ten behind GPT-6 Astra on FrontierSWE v2, and it sits around ninth on LMArena’s WebDev board. Bloomberg reported that some of Google’s own staff found it weaker at real coding work than the scores suggest. Google’s response was that it would be “inaccurate” to say Gemini 4 is underperforming, which you only need to say when people think it is.

It’s benchmaxxed as fuck. It’s tuned to post numbers for a launch blog, and it gets a lot less impressive once you hand it something messy. Google published the transformer paper in 2017 and has had more compute and more money than anyone since. It’s still chasing Anthropic and OpenAI.

And Apple had a choice. Mark Gurman reported that Apple talked to Anthropic about powering the new Siri, and that Claude was the front-runner in mid-2025. The deal died because Anthropic wanted several billion dollars a year, with the price doubling every year for three years. Google reportedly came in around a billion a year, so Apple took the cheaper model for the product its whole AI comeback rests on. Gurman also says Apple’s own internal development runs on Anthropic. The engineers in Cupertino get Claude and the people buying HomeView get Gemini.

Then Apple buried it inside Siri. You’d never guess from the HomeView page that there’s a large language model in there. Siri AI is in beta, English only and not available in the EU. Anything involving your personal stuff needs a paired iPhone or iPad with Apple Intelligence switched on, and HomeView wants you to unlock it with your phone every so often. The personality settings change how expressive Siri sounds and how fast it talks, and that’s the lot. There’s no App Store either, which is why Engadget compared it to CarPlay.

The only character on the whole screen is a Snoopy face that changes with the weather. The Apple Watch has had a Snoopy face since watchOS 10 in 2023.

I wanted a little character that lives on the screen. You pick one when you set the thing up and make it yours, whether that’s a cat, your Memoji or some lumpy blob your 9yo drew with their finger. Name it whatever you like. It hangs out in the corner while you cook, and there’s a good model behind it when you talk to it.

HomeView already recognises everyone in the house, so the character would too. It’d greet you differently from your partner and remember you asked about lasagne three Sundays in a row. When the 14yo wanders into the kitchen at 11pm on a school night it might have something to say about that (or not, if you’ve told it to mind its business). After a few months it’d know the house.

None of this needs new tech. ChatGPT, Gemini and Claude all keep memory now, and Character.AI has a whole business built on people talking to characters for hours. Tamagotchi sold tens of millions with a monochrome screen and three buttons. Kids get attached to anything with a face, and so do plenty of adults (I’m looking at everyone who still checks on their Animal Crossing villagers).

Apple even has the parts lying around. Memoji has been on the iPhone since 2018, it licenses Snoopy, and the face recognition is already in the box. Drawing a cat that talks would have been the easy bit.

They didn’t do it because Apple treats AI as something that should disappear into the OS. You can just about get away with that on a phone. HomeView’s whole job is sitting in your kitchen waiting to be talked to, and hiding the AI on that makes no sense to me.

I’m also not sure I’d want Siri to have a face, given its record. Siri launched with the iPhone 4S in October 2011, so it turns 15 this month, and it’s been bad for most of that time. It mishears you constantly and loves replying “here’s what I found on the web” when you asked it to do something. Most people use it for timers and the weather because that’s all it’s ever done reliably.

At WWDC in June 2024, Apple promised a personal Siri that understood your messages and your routines, and it ran TV ads for it. In March 2025 it admitted the features would take “longer than we thought” and pulled the ad. Bloomberg reported the home hub was held back too, because it needed the new Siri. So HomeView sat in a drawer for an extra year and a half while Apple worked out it couldn’t build the model itself and went shopping for the cheapest one it could get.

Spotlight on the Mac is in a similar state. Type “Calendar” and often Calendar isn’t the top result, and sometimes it isn’t there at all. Brent Simmons wrote in February that Spotlight had got terrible at launching apps after years of being good at it. The fixes people swap on forums are rebuilding the index, force quitting corespotlightd in Activity Monitor, or adding your whole disk to the privacy list and taking it off again. Raycast and Alfred have built businesses on Spotlight being bad. Apple can’t get its own launcher to find its own Calendar app, and now it wants its assistant in the middle of your house.

The timing makes it worse. OpenAI’s first device, designed by Jony Ive, won’t ship before the end of February 2027 according to OpenAI’s own court filing, and reports say it’s screenless and voice-first. Google has Gemini for Home on a US$99 speaker and Amazon has Alexa+. Nobody has made an AI device people love yet. For a few months Apple had the brand people already let into their homes and enough money to buy the best model on the market, and it went with the discount option.

OpenAI only does the model. Whatever it ships next year will be built around ChatGPT and everything it already remembers about hundreds of millions of people, and I think that’s the device that ends up defining this category.

If anyone at Apple is reading, there’s still time to call Anthropic back and put a cat on that screen before February.