When ChatGPT took off, everyone wanted to learn prompt engineering. People shared templates and debated personas, XML, markdown, and chain-of-thought. Entire businesses emerged around writing the perfect prompt.

We’ve been optimizing the wrong thing. The biggest productivity gains from LLMs come from learning to externalize judgment. Better prompts barely move the needle by comparison.

I’ll know it when I see it

Imagine you hire a designer. You ask for a “premium” website. A week later they show you a design, and it’s not premium enough.

”What do you mean by premium?” they ask. “It just doesn’t feel premium,” you say. Neither of you is wrong. The problem is that “premium” exists only in your head.

Your designer has one mental model. You have another. The project becomes a series of guesses until you eventually land somewhere you’re both happy with. This is an old problem. It’s been part of every creative profession for decades.

What’s changed is that now everyone has access to an LLM. Instead of occasionally hiring a designer or a consultant, we commission bespoke work dozens of times a day. LLMs didn’t invent the problem of communicating taste. They democratized it.

The mirror we didn’t ask for

One of the most common complaints about LLMs is that they don’t get it. Ask them to make something more engaging, more premium. Sometimes the result improves, sometimes it gets worse.

It’s tempting to conclude that the model isn’t smart enough. But after months of working with LLMs, I’ve come to a different conclusion. LLMs don’t eliminate the need for clear thinking. They make it impossible to hide its absence.

Humans are remarkably good at filling in the gaps. A designer asks follow-up questions, an engineer infers intent. They compensate for our ambiguity. LLMs don’t. They faithfully expose it.

That can be frustrating. It can also be one of the fastest feedback loops you’ll ever get for improving your own thinking.

From intuition to judgment

Many of us operate on intuition. We know good work when we see it. We know when a blog post feels too long or when a UI feels cluttered. But knowing isn’t the same as explaining.

The moment you try to tell an LLM what you want, you discover how much of your expertise is implicit. “Make it more engaging.” How? “Make it more premium.” Compared to what?

The LLM has plenty of taste. The problem is that your taste exists as intuition rather than language. The real skill is translating that intuition into explicit judgment.

Judgment outlasts the prompt

A great prompt can produce a great result. But prompts are artifacts. Tomorrow you’ll face a different task and write a different prompt. Judgment transfers.

Say you’ve spent time understanding what makes an article compelling. You might land on principles like these:

• Challenge conventional wisdom early.
• Introduce concrete examples before abstractions.
• Every section should teach something new.
• Leave readers with one idea they can’t stop thinking about.

Those principles don’t belong to one prompt. They become part of how you think. Your prompts improve because your judgment improved. The prompt is just one expression of it.

Why this matters

The people who get the most out of LLMs treat every interaction as an experiment. Instead of asking why the model got it wrong, they ask what they failed to communicate.

Instead of saying “make it better,” they ask themselves: I’ll know it’s better when… That shift changes everything. You’re no longer iterating randomly. Instead, you’re refining your own understanding of quality.

The next literacy

For years, computers struggled to produce. Now they produce endlessly. The bottleneck has moved from generating work to defining what good looks like.

That requires a skill most of us were never taught, because we could usually rely on other people to bridge the gap. Today, you’re often both the customer and the expert. There’s no one in the middle to translate vague aspirations into concrete decisions.

The LLM can’t read your mind. So you’re forced to do something uncomfortable: make your standards explicit. This is the next literacy: learning how to externalize judgment.

Once you can explain what good looks like, prompts become easier and evals become possible. Iteration becomes systematic. And AI starts scaling something far more valuable than output. It starts scaling your thinking.