People. Process. Product.

The interface is a conversation now. You type what you want, and increasingly you get it.

I've spent my career navigating this. People swear they have a clear vision but can't answer any follow-up questions. The stakeholder who says "I'll know it when I see it." That's not vision. That's deferred judgment. A polite way of saying: I don't know what I want, but I'm not ready to admit it.

AI didn't create this problem. It made it impossible to ignore. I love this! Hooray confrontation!

Ask most people why they love their Porsche. "It's just beautiful." Push them and they stall. That's why badge brands exist. Porsche manufactures the vocabulary for you. Wine does it even better. The tasting room gives you the heritage, the cinnamon notes, the founder's ritual. Now you have language.

There's no equivalent for knowing what you want from AI. No tasting room where someone walks you through the difference between "make this page look better" and "increase the visual hierarchy of the primary CTA while reducing cognitive load in the secondary navigation." Same desire. Wildly different precision.

The answer isn't "prompt engineering." It's philosophical. The same skill a good lawyer has, or a good designer: taking something foggy and making it specific. Sitting in uncertainty and poking at it until something solid emerges.

Most people don't know that they don't know. That's the real problem.

We're spending billions making AI smarter. We're spending nothing making humans better at knowing what they want.

You see a talking head on TV explaining how AI will replace all your jobs. It's always executives and investors. Never makers. Why?

Because there is money to be made in the idea — let me say that again — the idea of decoupling capital from labor is profitable. Not implementation. Not execution. The idea.

Mention that your new AI thing will replace labor and your startup valuation goes up. You can raise money on better terms. You literally profit off the idea of the future. And the beauty of AI is that it's a literal black box! Few fully understands what's inside, which means nobody can call your bluff!

AI is the most incredible thing I've encountered in my career. It is genuinely unreal… at times. But the ability for humans to extract value from AI is now the limiting factor, not AI itself. And that's the part nobody on TV wants to talk about.

I remember the first dot-com bubble selling the same pitch: leverage labor, scale without headcount. That was valuable until it wasn't. This is not that. We are legitimately decoupling labor from capital now. That is valuable. The question is who's building the real thing and who's just selling the idea of it.

Find the real ones.

There's a Simpsons episode where Homer gets to design his own car. His brother Herb runs an auto company and gives Homer total creative control. Engineers indulge every request: bubble dome, three horns that play La Cucaracha…

The result bankrupts the company. Nobody wants one. But Homer is exceptionally proud of it, because he made it.

AI is now delivering tens of thousands of Homer cars a day. Each one unique. Each owner beaming.

Look what I built!

It is exceptionally impressive! Someone who's never built a car before just rolled one off the assembly line.

What happens when it breaks down? The owner pops the hood and finds a rats' nest of wiring that they never even considered during implementation.

What happens when the requirements change? The Homer car wasn't built for any of that. It was built for the thrill of the first drive.

Welcome to the current state of vibe coding.

We're in the joy-ride phase. La Cucaracha blasting, wind in your hair. What comes next is hundreds of thousands of Homer cars, hoods up, hazard lights on, owners on the side of the road wondering what went wrong.

Just because you can does not mean that you should.

We're heading into a future of hybrid documents. Part human, part machine. You write the brief, the AI fills in the research. You confirm the diagnosis, the model suggests the treatment. This is already happening at scale. There's no convention for telling the two apart.

The obvious move is icons. Little robot emoji next to AI stuff, checkmarks next to human stuff. But a dense report covered in those badges is visually noisy.

So here's the pattern: inference italics.

Inferred content goes italic. Confirmed content stays roman. That's it. It works because italics already carry the right connotation: aside, different voice, not quite the main speaker.

They're the web's most underused typographic tool! Bold does all the heavy lifting while italics just sit there. This gives them a proper job. Here's the bonus: the <i> tag in HTML5 means "alternate voice or mood." That's literally this use case! The infrastructure already exists.

Is it perfect? No. Screen readers need a fallback. Humans might want italics for emphasis. The bigger point is we need something, and the simplest signal usually wins. Conventions don't spread until they have a name.

I'm calling it inference italics.

Take my 1 question survey to let me know!

I do. Not all the time, but enough of the time.