AI has moved into every phone. Now what?

—Even Apple is renting Google's brain: once "do you have AI" stops being the question, the question becomes "what do you do with it"

Yesterday's Apple WWDC did something that would have been hard to imagine three years ago.

It tore Siri down and rebuilt it entirely, running on acustom, 1.2-trillion-parameter Google Gemini . At the same time, it opened a multi-AI "Extensions" framework, letting users choose Claude inside the iPhone too.

In other words: from today, a few of the world's most powerful brains live inside every iPhone.

Everyone's arguing about whether "Apple AI has failed" and "why they're renting Google's instead of building their own." All fair—but what I see is something quieter, and more important.

AI access has officially become a utility.

Even Apple is renting

Let's start with the fact most people gloss over in a single line, but that I think is the whole point:Apple can't build it itself.

The richest company in the world, the best at making chips, with well over a billion devices in its hands, has conceded defeat on the one thing of "owning a brain strong enough." It chose torent Google's.

That carries an enormous signal.

Because if even Apple—the company with the device, the strongest moat in history—decides not to grow intelligence in-house but to buy it, then it's telling everyone:The model itself is not the moat. It's a raw material you procure on the market, like electricity, like tap water, like cloud data centers.

In my year-plus of building things, this has been my deepest realization too. I build saomin.tw/me, build customer-service AI, build tarot readings—and underneath it all sits a large model. But I've never felt that "I used Claude" was anything special—because you can use it, they can use it, and now even your aunt's iPhone can use it.

When everyone has something, it stops being your advantage. It's the floor, not the ceiling.

When intelligence becomes a utility, what's scarce?

The nature of a utility is this:It makes "having it or not" a non-question, pushing the question one step further down the line.

A hundred years ago, whether your home had electricity was a measure of wealth. Today electricity is taken for granted; the difference is what you do with it—run a factory, or leave a bulb on and stare into space.

AI is heading down the same path. Before long, asking "do you use AI" will be as meaningless as asking "does your home have electricity." Every phone has Claude, has Gemini; the questions are all equally smart, the answers all equally fluent.

So where's the difference then?

The difference is inwhat you bring to it that the general-purpose brain doesn't have.

I can think of three things:

First, your data. The model has read the whole world, but it hasn't read you. It doesn't know every customer-service conversation your company has had over the past three years, the temperament of a particular client, or the words you've written over twenty years. With me.saomin, I'm not building "an AI that can chat"—that's worthless, everyone has it—I'm building "a container loaded withme." What's valuable was never the model; it's the person you pour into it.

Second, your judgment. General AI will hand you a thousand workable options, but it won't tell you which one is right. "Which one is right" takes a whole set of taste and standards that are hard to articulate but unmistakably yours. However strong the brain in your phone, it doesn't know howthis particular callshould be made—because it isn't you, isn't in your situation, doesn't bear your consequences.

Third, the vertical you stand on. The general-purpose brain knows a little about everything, but it doesn't know "how to reply to a player who's furious about a specific bug on a specific map in a specific game." Depth always beats breadth—as long as you're willing to hold a spot narrow enough and deep enough.

But don't make too much of this

I have to honestly tap the brakes, because news like "AI in your pocket" is the easiest kind to get overexcited about.

First, access isn't outcome. Having one more genius in your pocket doesn't mean your life gets better. A while back I watched someone use AI to spin up sixteen projects at once—and finish none of them. A higher ceiling on your tools doesn't mean a higher ceiling on your output—what stalls people was never "can I," but "do I know what to do" and "did I see it through." However strong the brain in your phone, it waits for you to speak; if you don't know what to ask, it's just a more expensive standby screen.

Second, it isn't everywhere. Apple said so itself: the new Siri's AI is not in Europe, not in China—regulation is in the way. A lot of what I build is aimed at users in China, and for them, "Claude inside the iPhone" simply doesn't exist. So those fantasies of "the platform has it built in for you" are empty in many markets; you still have to wire AI into your service yourself. For me that's actually a good thing: where the platform can't reach, what you build yourself has room.

Third, built-in convenience usually costs you the right to decide. When AI is what the platform picks and bundles for you, you use the one version, the one flavor, the one boundary it lets you use. I'd rather take the trouble—wire it myself, tune it myself, decide myself what it remembers and what it's not allowed to say—because it's my service, not Apple's.

In closing

On the surface, that WWDC was Apple conceding, Google winning a round, and Claude squeezing into a phone.

But for someone using these tools to build things, the real message is just one line:Intelligence is becoming a utility, and a utility won't make you win—it just pulls everyone to the same starting line.

After the starting line, the ones who can run are those whobrought something with them—their own data, their own judgment, the narrow patch of ground they've guarded for a long time.

The brain in your pocket now belongs to everyone.

The only difference left is: what you aim it at.

News source (WWDC 2026): - WWDC 2026: Everything announced on Siri AI, iOS 27, Apple Intelligence and more — TechCrunch - Apple WWDC 2026 live updates — CNBC - AI News Today — June 8, 2026 — Build Fast with AI

Further reading: - When AI starts building itself, what's valuable is no longer "being able to do it" - It wants to be AI's upstream; I just want to leave behind an echo - How to build an LLM that's all your own