You Don't Need to Understand AI. You Need to Know What It's Actually Doing.

There's a version of this essay that starts with "AI is transforming everything."

This is not that version.

If you're a designer or creative professional who's spent the last two years watching the AI conversation happen around you, with a mix of curiosity, skepticism, and some quiet anxiety, this is for you.

Not the version of you who needs to be convinced AI is real. You know it is. You've seen the outputs. You've used a tool or two.

This is for the version of you who still isn't sure what to actually do with it, and suspects that most of what's being written about AI is either too technical to be useful or too enthusiastic to be honest.

That suspicion is probably right.

Before getting into AI, there's one useful idea about understanding itself: understanding is not binary.

Knowing what something is, being able to repeat an explanation, or getting a good result once does not mean you deeply understand it. Real understanding has levels. The more you understand something, the better you can explain it, test its limits, connect it to other ideas, ask better questions, and change how you act because of it.

That matters with AI. The question is not simply, "Do I understand AI?" It is: "Do I understand it well enough to make better decisions about my work?"

What AI actually is, without the mythology

Most explanations of AI start with how it works: neural networks, training data, parameters. It's accurate and almost completely useless for a designer trying to decide whether to use it on Tuesday.

Here's a more useful frame:

AI is a pattern-completion machine. Trained on human output. Very good at the middle of the bell curve. Genuinely surprising at the edges, in both directions.

When you give it a prompt, it's not thinking the way you are. It is predicting what a reasonable continuation looks like based on patterns it has learned.

Sometimes that prediction is extraordinary. More often, it is competent. Occasionally, it is confidently wrong.

Understanding this changes how you use it.

You stop asking: "Can AI do this?"

You start asking: "Is this task closer to pattern-completion or judgment?"

AI is very good at pattern-completion. Judgment is harder because judgment depends on context, stakes, taste, relationships, and things that often are not visible in the prompt.

That distinction, pattern versus judgment, is still one of the most useful ways to think about AI.

There's another principle worth keeping in mind too.

Shreyas Doshi calls it the Antithesis Principle: when you learn something useful about how people or systems work, look at it in both directions.

If AI can generate convincing answers quickly, that gives you a useful capability. But it also gives you a warning: don't confuse confidence with correctness.

If AI can make you much faster, great. But ask: faster at what?

If AI removes friction from thinking, useful. But not every kind of friction should disappear. Sometimes the struggle is part of understanding.

The same capability can be both a tool and a warning.

The mistake most designers make on day one

They start with the wrong question.

The wrong question is: "What can I use AI for?"

It sounds reasonable. It's actually too open. When the answer is close to "almost anything," you end up scattered, trying a tool here, a prompt there, never building real fluency.

The better question is:

"What in my current work is repetitive, frustrating, or takes longer than it should, and doesn't require my specific judgment to be good?"

For most designers and creative professionals, that might include writing first-draft briefs, generating copy variations, resizing assets, summarizing research, organizing notes, or describing what a design does for handoffs and proposals.

These are not usually the interesting parts of the work. They are the scaffolding. And scaffolding is exactly where AI can help.

The interesting parts are the creative call, the direction, the taste, the decision that something technically good still feels wrong.

That is judgment.

No one hired you for a statistically average output. They hired you for your judgment applied to their problem.

What "AI basics" actually means in practice

Everyone who writes about AI fundamentals eventually gives you a list of tools. ChatGPT, Claude, Midjourney, Runway.

Tool lists age badly.

What lasts longer is the mental model for how to work with these systems well.

A few fundamentals hold across almost every AI tool.

1. Specificity beats cleverness. Be clear about what you want, who it is for, what context matters, what format you need, and what should be avoided.

2. The first output is a draft, not an answer. The people getting real value from AI push back, refine, add context, compare options, and keep working with the system.

3. You need to know enough to edit. AI becomes more useful when you can judge the quality of what comes back. Expertise is not made redundant by AI. It is often what makes AI useful.

4. Your input quality shapes your output quality. "Make it better" is weak. "The hierarchy is failing because the eye goes to the subtitle first" is useful. Clear thinking produces better interaction with AI.

The honest version of where this goes

I run a design subscription company, Hubee, operating in Brazil and MENA. We've been integrating AI into our workflow for over a year, not as an experiment, but as a business decision.

Here's what I've seen.

AI didn't replace designers. It changed what they spend their time on.

Work that was administrative, repetitive, or structural started getting absorbed. Drafting, organizing, researching, summarizing, creating variations.

The work that depended more on craft, taste, context, and judgment stayed much more human.

The designers who adapted fastest weren't necessarily the most technical. They were the ones who had a clear sense of what they were actually good at, and therefore a clear sense of what they were willing to hand off.

That clarity matters.

AI can make you faster, but the more important question is still:

Faster at what?

What to do after reading this

Not "sign up for these five tools."

Not "start your AI journey today."

Do one thing.

Write down the three tasks in your current work that take the most time relative to the value they produce.

Then ask:

Which part is pattern? Which part requires my judgment? What could I hand off? What should stay with me?

And one last question:

If this capability exists now, should I still be doing the work the same way?

That is a much better place to start.

Because the goal is not to use more AI.

It is to understand it well enough to know where it belongs, where it does not, and what it should change about the way you work.