My iPhone 18 Pro Max arrived Friday. I’ve spent the weekend and this morning’s school run living with Siri AI, and I want to give you the honest version, because I wrote about this launch in August before I could touch it and I owe you the follow-up.
Short version: it’s the biggest jump Siri has ever made, the architecture is the story, and the gaps are exactly where I expected them.
The part I care about most
In August I argued the two-layer design was what mattered. Having used it, I’d sharpen that: the layer that matters is the one on the chip.
The A20 Pro has a dual 16-core Neural Engine, and most of what I ask Siri to do all day runs right there. Reading my calendar, drafting a reply, pulling a meeting summary, building a playlist. No round trip to a data center. Faster, obviously. But also quieter in a way I find myself appreciating: a request that never leaves the phone has no server to spin up, no cooling to pay for, and no copy sitting anywhere.
I want to be precise here because the precise version is still impressive. Apple used Gemini to train the smaller models that live on the device. When a request exceeds what the phone can handle, it routes to Apple’s Private Cloud Compute with personal identifiers stripped, and for the heaviest lifting that means Google’s infrastructure under Apple’s controls. So it’s a hybrid. But the default is local, and after years of assistants that phoned home for everything, that default is the whole point. It’s the difference between an assistant and a surveillance product with good manners.
What it’s actually good at
Conversation, finally. You can talk to it like a person and it keeps up. Ask a follow-up, change your mind mid-sentence, and it stays with you.
Mail and Calendar are where it earns its keep. I run multiple Google accounts plus iCloud, and it sees all of them without a single authentication screen. If it’s connected on the phone, Siri can use it. That sounds small until you remember every other assistant makes you rebuild your life inside its app first.
CarPlay is the surprise. Navigation, creating calendar events by voice, having my next meeting read back to me across three calendars. Zero errors across the weekend. This is the use case where a mistake actually matters, and it didn’t make one.
Apple Music playlist generation is fun and it works. The standalone Siri AI app is smart: it logs your voice conversations from CarPlay as text, so the thing you asked for on the freeway is waiting for you when you park.
And the real-world test. My youngest wanted to know about snow this morning on the drive to school. When does it snow, where does it snow, why doesn’t it snow here, does it snow on the moon. Rapid fire, no pauses, the way kids ask. Siri handled every one of them without missing a beat. I’ve been in this industry 25 years and that was the moment I thought: okay, this is different.
What’s missing
Third-party apps, mostly. Developers have to adopt Apple’s App Intents framework before Siri can reach into their apps, and the ecosystem hasn’t caught up. Granola, which I use for every meeting, isn’t there yet. Neither is Google Workspace, so Docs and Sheets are invisible to it. I have a hunch Google won’t rush to integrate, given how much it wants you inside Gemini instead, and that’s a real limitation for anyone who lives in Workspace.
The cloud rollover for complex questions currently goes only to ChatGPT. Claude and Gemini are announced and coming, but as of this weekend the menu has one item. The voice-change disclosure I wrote about in August is real and I like it; I just want more voices to hear.
And it’s a beta, on a waitlist, English only, not available in the EU or China. Apple is being careful, which I’d rather see than the alternative.
The verdict
Siri AI is going to become my daily driver for the things phones are actually for: managing the device, handling calendars and mail, answering the questions that come up in a car. It is not going to replace a full frontier chatbot for deep work, and it isn’t trying to.
For anyone building products in this space, three things to watch. App Intents is now the gate to a billion devices; if you’re not building for it, you’re invisible to the most-used assistant on earth. The on-device default resets the privacy conversation for enterprise buyers, and your competitors will be asked why their AI can’t say the same. And the Extensions framework is a distribution channel that didn’t exist a month ago.
Apple will push developers hard on all three. Based on three days, they’ve earned the leverage.