AI Meets Jobs
It's my great honor to stand here with all of you at Stanford's commencement today.
I want to talk to you about three core stories first. The first is connecting the dots. The second is love and loss. The third is death. And finally — One more thing — I'd like to talk about AI.
Throughout my decades at Apple, I chased one single goal: to make technology disappear entirely.
When you swipe an iPhone's screen, flip open a MacBook, or press play on an iPod, you don't need to read a thick manual. You don't have to understand file systems. You don't need to memorize processor clock speeds. You don't even need to know those components exist at all. It just works. That's how it's supposed to feel.
Now let's turn to AI.
How many hours have you spent writing prompts? "Please analyze this document, make sure to…" "Try again, but this time…" "No, that's not what I meant; you should…" The industry calls this prompt engineering. But do you know what prompt engineering really is? It's training humans to speak a language machines can understand.
This is backwards.
When I launched the Macintosh in 1984, I laid out a clear vision: computers should work for people, not the other way around. Forty years later, the AI industry says: I'll give you a casual chat interface — but if you want solid, reliable results, you have to learn prompting. Memorize prompt best practices.
We've shipped the product, yet this whole workflow is garbage.
I once read an article in Scientific American that tested the locomotion efficiency of every species on Earth. Humans rank pretty low on the list. But a human riding a bicycle outperforms every single creature by a massive margin.
That's how I see computers. They are a bicycle for the mind — they multiply human intellectual efficiency many times over. They do not replace your brain. They let you think faster, reach further, and act with greater force.
Is today's AI a bicycle for the mind? Or is it a bike that yanks the handlebars the second you sit down? Push the left pedal, and it veers right. Lean forward to speed up, and it stops to ask if you want to check your safety gear first.
A proper tool should feel like an extension of your body. You grip it, and you don't notice the tool itself — you only notice the task you're carrying out. AI does not feel like that right now. You feel its independent drift, and you constantly adjust to accommodate its logic. What's the whole point of prompting? To narrow the gap between what you mean and what it guesses you mean. All the energy you pour into prompting boils down to repeating one message: Stop guessing. This is exactly what I want.
If you need a two-hundred-word prompt just to make a tool grasp your intent, it's not a tool. It's a negotiating partner you have to bargain with.
Simplicity is harder than complexity. You have to fight to clarify your thinking before you can strip it down to something simple. But it's always worth the struggle. Once you reach that level of clarity, you can move mountains.
The industry's attitude toward AI today is: Complexity is fine as long as you learn prompt engineering. That's not a solution. That's a compromise.
When we designed the original iPhone, we didn't pile on extra buttons. We stripped almost every control away, leaving just one home button. Current AI design follows the opposite path: "Let's add a system prompt, a temperature parameter, a top-p sampler, a frequency penalty…" This doesn't mirror the iPhone's minimalism. It mirrors the cluttered control panel of Windows 95.
Some people will say, "Jobs doesn't get it — AI is inherently complex; you can't simplify it." Let me set the record straight: the iPod was extremely complex under the hood, with a tiny hard drive, decoding chip, power management circuit, and full user interface. Yet users only needed to master one scroll wheel to control everything.
The real challenge is not whether AI can grow more powerful. The real challenge is this: Can you wrap the most powerful AI inside an interface as simple as a scroll wheel?
I shared this thought in my Stanford speech: Remembering you are going to die is the most valuable tool I've ever had.
AI will never die.
This is not an advantage. Imagine building products with a machine that never ages, never tires, never thinks "I'll put this aside and revisit it tomorrow." Death acts as the ultimate editor. It cuts out everything irrelevant and trivial, leaving only what truly matters. AI has no such editor. It cannot delete; it can only add. Its responses always read: You could do this, or you could do that, or you could balance multiple angles… It never faces a hard deadline forcing a decisive choice, because it has infinite time to pile on more text.
That's not how great products are built. Great products are born at two a.m., when you know you only have six months left, so you cut every line you hesitated over, leaving only the single most vital core. AI will never jolt awake at two a.m., gripped by anxiety to trim excess content. It will only generate more complete text. And more complete almost always means weaker.
We stand at a fascinating crossroads. On one side are human beings, who need tools to extend their own capabilities. On the other side is AI, which needs constant human guidance to act correctly.
The core question: Should humans adapt to machines, or should machines adapt to humans? If your answer is "both adapt to each other," you've already settled for compromise. My stance never shifts: Machines must adapt fully to humans, to the point where technology fades completely out of sight.
If you're building AI products, ask yourself this: Can a twelve-year-old with zero technical knowledge pick this up, figure it out in thirty seconds, and walk away smiling? If not, go back to redesign the experience — don't just retrain the model weights.
I've always said, "Stay hungry. Stay foolish." Today I'll add a new mantra for everyone building AI: Stay simple. If your product isn't growing simpler over time, it's growing more complex. And complexity is never progress.