Apple's flagship AI assistant, unveiled in June, runs on a model built by Google.

Apple's flagship AI assistant, unveiled in June, runs on a model built by Google. Apple pays Google to license Gemini as the intelligence behind the new Siri AI. Two companies that compete on almost everything else are now, at the model layer, running the same engine.

This is the sixth in a short series on the different types of agents shaping how, and whether, brands get chosen. Every category so far, chat apps, marketplaces, third-party apps, browsers, seller tools, lives inside something a shopper has to open. This one sits underneath all of them e.g. the operating system itself, intercepting intent before any app gets involved.

Apple's Siri AI, announced at WWDC in June, brings multi-turn conversation, system-wide awareness of what's happening across apps, and genuine agentic action; its Passwords app can now visit sites and update credentials on its own. It's not available in the EU or China yet, still working through regulatory approval in both.

Android's approach is more crowded. Samsung's Galaxy S26, launched in February, ships with three separate agents to choose from on the same device, Bixby, Gemini and Perplexity. Gemini can already handle multi-step tasks across apps, Google and Samsung cite preparing a grocery cart or reordering a regular meal as examples, with the user reviewing before anything completes. Google still labels this experimental.

Microsoft's Copilot is the most established of the three, baked into Windows and Office, though it's drawn steady criticism for being persistent to the point of intrusive rather than for what it can actually do commercially yet.

None of this is transacting at scale today. It's the least mature category in this series, gated by region, still marked experimental, mostly limited to task assistance rather than purchases. But it's worth watching for a specific reason: if an OS-level agent is the one assembling a grocery cart across several apps, which brand actually won that moment stops being something visible in any single app's own analytics. The competition moves a layer up, to whichever assistant gets chosen as the default on the device itself, before your app, your feed or your ad ever gets a look in.

Apple licensing Gemini is the detail worth sitting with. The two biggest phone platforms may end up reasoning with very similar underlying models. What will actually differ between them is the permission and data-access architecture each builds around it, what the assistant is allowed to see and do on a person's behalf. That's the thing worth tracking here, not which model is technically smarter.