Prompt Engineering Isn't Dead. Your Requirements Just Got Promoted.

Prompt engineering is one of those things that, when we first started working with LLMs, you had to do a lot of heavy lifting to make sure ChatGPT or Claude was doing what you wanted it to do.

You'd say, "You are a copywriting expert." Or, "You are a social media guru." Whatever it was, you had this structure. And so we started calling it prompt engineering. Like, hey, there's a discipline here. There's an engineering component to this.

We'd write really large checklists of things to cover in the prompt. We got good at it.

And then the models got really smart.

When the Models Stopped Needing Training Wheels

Over the last 6 to 12 months, depending on what you're doing, you've probably discovered something: you don't need to worry so much about it anymore.

You don't have to say "you're a copywriting expert" when you're asking about a copywriting effort. The models are smart enough to go, "Oh, I know a lot about copywriting. I can put some work into it."

So then we said, "Oh, okay. Prompt engineering is dead."

And I get it. If what you're really talking about is the art and structure of trying to limit and lock in what the LLM does, yeah, that part of prompt engineering may not be the thing you need to do anymore.

Here's What They Got Wrong

But prompt engineering was never just about tricking or scoping or managing the LLM.

It was also about articulating what you care about. How you work. What kind of result you're looking for. What format you're looking for. Structure that you care about.

All of that is about your own expectations.

And when someone says "prompt engineering is dead," it's ridiculous. But your requirements aren't dead. They still exist and they're unique.

When your requirements are unique, you must spend energy and time articulating what it is you want and how you want it to work. That part never went away. It just got promoted.

The Skills-Only Trap

Now I know right now there's a big wave of people saying, "Oh, it's skills, it's skills, it's skills." And look, I created a product called Your Voice Profile and another one called Your Audience Segments. They're markdown files.

A guy bought Your Voice Profile the other day and said, "Oh, I bought your voice profile, but it should be a skill instead."

Here's the thing. The problem when you only say "I'm gonna only use skills, I'm gonna put everything into skills," is that you're locking into where and how you're gonna use it.

You may think, "No, no, no. I can make a skill that can be used in Claude and also in Codex and also in Gemini and also..." And sure. That's not what I'm talking about.

Why Portability Matters More Than You Think

The other day, I was using someone else's SaaS. Testing an AI-based SaaS that would help me write. It's been trained on all sorts of writing. But it has some background spots. You know, "put in your personal background, put in your profile."

And I just copied my entire voice profile and dropped it in.

Which you can do with a markdown file. I can't get my skill to suddenly show up in someone else's product. But I can get my voice profile in someone else's product by dropping it in.

We can put our expectations of how we want things working, the formatting, the things we want to have happen and things we don't want to have happen. Those are all just expectations. And being rigorous about that is critical.

What Prompt Engineering Actually Became

So when you hear "prompt engineering is dead," I want you to think of it differently.

The old prompt engineering? The one about tricking and scoping and managing the LLM? Yeah, maybe that's fading. The models are smart enough now that you don't need to hold their hand through every step.

But the new prompt engineering will never go away.

Learning how to articulate your goals and desires, your strategies, your intent, and your framing. What you care about and what you want to come out.

That's what makes the output valuable.

Embrace the Conversation

But let's be clear. The output isn't the most important part. Nor is the initial prompt. It's the conversation. The whole prompting process, the back and forth. That's the real value, not just because what comes out is great.

No, it's more than that.

What happens in the conversation is that you're getting more clear about what you want, how you want something to work, and how you want it to look.

The process of clarifying your own thinking, via the conversation, means you bring that level of intelligence now to your next conversation, whether it's to AI or to another colleague.

Here's a quick (less than 5 minute) example.

Single Prompt Engineering Is Dead

Maybe what we should have said all along is that single-shot prompt engineering really is dead. And dumb. And a waste of time.

The real work is the conversation, the multiple prompts (which you can't download from someone's "the last 7 prompts you need"), that help you articulate your unique requirements.

The models got smarter. Your requirements didn't get simpler. If anything, the fact that the models can do more means you need to be even more focused on what you want.

That's not dead. That's just getting started.