Siri, built partly on Google's Gemini, leads all three home devices Apple reportedly plans for October

In a shared kitchen, someone asks the countertop hub, a 6-inch square screen ^[2], to read the text that just arrived. Before Siri answers, the hub has to settle whose text it is. When voice or face is not enough, Bloomberg says, that person may have to confirm on their iPhone ^[7].

The hub is built for households where every adult already carries an iPhone. According to Bloomberg, Apple pictures several of them around one house, used to talk to each other, watch video and make FaceTime calls ^[9]. Each will likely need to stay plugged in ^[4].

Bloomberg's case for the hub, the HomePod mini and the fourth-generation Apple TV 4K rests on one feature, Siri AI ^[1][6]. Apple first promised that Siri at WWDC 2024 ^[3]. According to BGR, Apple managed basic AI features on its own ^[15]. The Siri these devices lead with shipped only after Apple paired exclusive Google Gemini models with its own Foundation Models and private cloud ^[15].

For teams building on Siri, the useful detail in the report is small. The new assistant can write shortcuts from plain language, home automations included ^[12]. The report does not describe developer access to the hub or any change to how apps and accessories pass actions to Siri.

Product teams tend to assume users will open the companion app, build a scene and name it. The hub asks them to say a sentence to a screen and let Siri assemble the automation ^[12]. When that automation switches on the wrong light, I'd expect the support ticket to reach the accessory maker, whichever mix of Apple and Google models misread the request ^[15].

The HomePod mini puts the same assistant in a speaker. BGR notes that Apple Intelligence currently requires an A17 Pro or M1 chip with 8GB of RAM ^[13], and that Apple raised the mini's price a few months ago ^[14].

The decision for Monday is a sort of every action a product exposes to Siri, on two tests: whether it depends on knowing which person asked, and whether a wrong guess is cheap to undo. Shared, cheap actions such as lights or a timer can run on voice alone. Personal, cheap ones, like one person's playlist, can ride on face or voice recognition, because a mismatch costs a skipped song. Shared, costly actions such as a door lock need confirmation whoever asks. For personal, costly ones, like reading someone's messages, the iPhone check Bloomberg describes belongs on the normal path ^[7]. I would ship that check and accept the friction, knowing each confirmation moves the task from the counter back to the phone.