Apple shipped a lot on September 9. A $1,999 folding phone. Two watches. A health app that wants to read your bloodwork. The thing worth your attention, though, is an argument that was everywhere within about six hours of the keynote.
The argument goes like this. A lot of ordinary AI tasks are going to become good enough. Once the assistant on your phone reliably finds what you need, makes the edit, handles the simple request, a smarter model somewhere else stops being worth paying for. Apple only has to be in the right place with the right information, which is a thing it already is. Ben Thompson made a careful version of this case back in June, and it’s the most serious defense of Apple anybody has written in two years.
I have sympathy for it. I also think it measures the wrong thing.
Good enough is a claim about the tasks we already know to ask for. It treats demand for intelligence as a fixed list that capability eventually catches up to and finishes. We have run that experiment before, with bandwidth. In 1997 nobody could explain why you would want a gigabit connection, because the honest pitch was that your email would arrive faster, and that wasn’t worth the money. Everything that eventually made the bandwidth worth having hadn’t been built yet. There was no bandwidth to build it for.
So the question this launch raises sits one step past whether everyday AI gets good enough. It’s what we’ll consider everyday after we’ve lived with it for two years. That question has a much larger bill attached to it.
That matters to you, not only to two enormous companies. Whoever is holding your work when the answer changes is the company you keep paying. The phone is the one-time purchase. The relationship is the recurring one. And Apple has already told us, in fine print almost nobody read on launch day, that it intends to charge for part of that relationship: certain features that rely on server-side models get daily usage limits, and expanded access will be available for a fee in the future. Apple didn’t print the number.
If you watched the video: that one covered the fight over which company you trust to run your life. This goes after the question underneath it. What happens to demand when intelligence gets cheap?
The article covers:
Why solving a task can create a bigger job. Finding your confirmation number is finished work. Rebooking the canceled flight is where demand actually starts.
How AI builds its own demand. Cheap models make small software worth writing, and that software is how people find out what the models are for.
Where Apple’s position is real and where it isn’t. Owning the device buys Apple the moment a need arises, which is an opportunity and not a settlement.
Why more usefulness doesn’t settle who gets paid. The value created by cheap capacity and the money kept by whoever supplies it are two different numbers.
When accumulated memory locks you in, and when it stops. A better model can rebuild context from material you already control. It can’t rebuild a reason nobody wrote down.
Start with what Apple actually shipped, because the fine print matters more than the hardware.
