Last week in Los Angeles, the day after Apple unveiled its touchscreen MacBook Pro, new Apple CEO John Ternus was blunt in an interview with Wired: "We don't ship AI for the sake of AI." If a feature works properly, he said, users shouldn't even need to care whether AI or conventional software produced the result. He also drew a line against the idea that AI agents will replace today's apps. "Making products is a deeply human and personal endeavor."
Ternus succeeded Tim Cook on September 1. The seat he inherited is not a comfortable one. Apple has been seen as lagging in the AI race for years, and the upgraded Siri only reached the market this summer. An executive in that position would typically promise to ship more AI features to make up lost ground, wouldn't they? He did the opposite and announced what Apple won't do.
The Test: Users Shouldn't Have to Know
What strikes me about this statement is less the "we won't" than the standard behind it. Saying users shouldn't need to be aware of whether AI was used means Apple won't use AI just to add a line to a feature list. That runs against the grain of how many companies now measure themselves, by how much AI they have put in. SK Telecom recently announced "AX Innovation 2.0," reporting that more than 4,000 employees have built agents themselves, 1,500 agents are running continuously, and 93.8% of staff say they support the change. A CEO-led task force is overhauling data access permissions, and each organization is running a three-month experiment to rebuild its entire workflow around AI. By speed and participation, it is impeccable execution.
I have no intention of faulting that speed. But the moment "how much did we put in" becomes the measure of success, organizations can easily slip into quietly inflating the numbers. Agent counts, participation, approval rates: all are meaningful metrics, yet how many organizations can also explain why they did not apply AI to a particular task? Ternus's remark cuts because it turns that question around.
Those Who Delivered Quietly Narrowed the Scope
Look at cases that produced numbers without much noise and this standard becomes visible. HD Hyundai Robotics applied Palantir's AI to exactly one task: resolving problems at industrial robot sites. Average handling time fell from 21.5 days to 7.9 days, a 63% reduction. There was no grand declaration of company-wide transformation or organizational redesign, only a focus on the single most repetitive task with the most accumulated data. I see these two cases as telling the same story. Ternus's "users shouldn't need to know" and HD Hyundai Robotics' "this one task only" connect at one point: both decide where AI belongs by usefulness, not by volume.
In nearly 30 years of examining enterprise technology strategy, I have seen this scene often. When a new technology arrives, organizations reliably split in two: those trying to show how quickly and extensively they have deployed it, and those who can explain why they have not yet applied it to a given task. The former looks impressive in presentation decks, but a year later the result is often a pile of unused features. I call this "proof fatigue," a state in which the effort spent demonstrating that AI was adopted exceeds the effort spent explaining what problem it actually solved.
It is worth asking yourself the same question. If you can list where your organization "put in" AI this year, can you describe just as clearly where you decided not to, and why? That is the first thing we check when we examine how an enterprise has structured its AI adoption. You might use our TechScan diagnostic to see where that standard currently stands.