Skip to article
The Design Office of Nick Sloggett
WorkWritingExperimentsResumeEmail
Back
Essay·Jan ’26·3 min read

The model is a supplier. The assistant is the product.

Apple’s next Siri runs on models based on Google’s Gemini. What that says about where an assistant’s design lives, and what Apple kept for itself.

Typeface

Apple’s next Siri will run on models based on Google’s Gemini. That sounds like Apple handing its assistant to a rival, but it says something more useful about AI products: the model is becoming a supplier, and the assistant people actually meet is a set of design decisions on top of it.

Is Siri using Gemini?

Yes, starting this year. On January 12, Apple and Google issued a joint statement: a multi-year collaboration under which the next generation of Apple Foundation Models will be based on Google’s Gemini models and cloud technology. Those models will power future Apple Intelligence features, including the more personalized Siri coming this year. Apple Intelligence will keep running on Apple devices and Private Cloud Compute. No financial terms were given, though Bloomberg reported in November that Apple would pay about $1 billion a year for a custom Gemini model.

This is the Siri Apple showed at WWDC in 2024, with personal context and actions in and across apps. In March, Apple said it would take longer than expected. The latest reporting points to iOS 26.4, with a beta next month.

What Apple kept

Read the statement for what stays with Apple. The models run on Apple’s devices and on its own private cloud. The statement doesn’t spell out the rest, and it doesn’t need to. Apple still decides what Siri can see on your phone, which apps it can act in, when it interrupts and when it stays quiet, how it asks before doing something, and what it sounds like. Gemini supplies the capability. Apple supplies everything a person would call Siri.

That split was always there. It’s just visible now.

Where an assistant’s design lives

Most of what people judge an assistant on has little to do with how smart the model is. Who is talking to it, and what is it allowed to show them? A speaker in a kitchen might be heard by a child, a partner or a guest. When should it speak first, if ever? What does it say when it doesn’t know, and how does it ask before it acts?

The AI gadgets of 2024 found this out in their reviews. They were sold as new paradigms and judged on old jobs: status, feedback, reliability, undo. None of those questions change when the model underneath changes. That’s why they’re the durable part of the product. A better model makes the answers more capable. It doesn’t answer them.

Google is living this too. In December it pushed the switch from Assistant to Gemini on phones into 2026, saying it wanted a seamless transition. That’s a design problem more than a model problem: moving people’s habits, routines and trust from one assistant to another.

What this means for teams building on someone else’s model

Most companies building AI features are in Apple’s position now, without the leverage. The model is rented. The product is everything you decide around it.

So spend design time where it compounds:

  1. What the assistant can see, and how a person can tell.
  2. What it can do on its own, what it has to ask about, and what it can never do.
  3. How it behaves when it’s wrong, and how a person takes an action back.
  4. Its voice: when it speaks, how much, and in whose name.

Swap the model next year and all four still hold. That’s the part of the product that’s actually yours.

Previous← Designing for operatorsNextGlasses need a smaller default →

© 2026 Nick Sloggett

EmailLinkedIn (opens in new tab)ResumeRSS