Prompting Is Now Briefing

I've spent a ridiculous amount of time thinking about prompts.

I've written them, taught them, sold them, built prompt packs around them, and made a pretty decent amount of money doing it. At one point, "prompt engineering" was a completely reasonable description of what we were doing. The models were brittle enough that how you phrased something could materially change whether you got something useful or a pile of nonsense.

Those prompt packs are all free now. They're still in the store at RichWashburn.com. You technically have to go through the store process, but they cost nothing. I've also put newer versions of them, mostly as Markdown files, in the Files area at CafeWashburn.com. They're useful educational artifacts. They show how we learned to communicate with these systems.

But I don't think prompting is where the value is anymore.

In fact, for getting real work done, I think the whole idea of "prompting" is rapidly becoming the wrong mental model. The better word is briefing. And Astra is making that painfully obvious.

For years, we've gone through this progression of increasingly fancy terminology. Prompt engineering. Context engineering. Intent engineering.

Cognitive architecture. All useful concepts. All technically defensible. And all increasingly unnecessary for the average person trying to figure out how the hell they're supposed to use AI.

If I walk up to somebody who doesn't live inside this stuff and say, "You need to improve your context engineering," I can practically watch their eyes glaze over. It sounds like a discipline they haven't studied.

It sounds technical. Worse, it sounds like I'm telling them there's some secret language they need to learn before they're allowed to use the machine properly.

There isn't. Brief the damn thing. That's it.

Think About How You Give Work to a Person

I've been explaining AI this way for a while now: treat it like an intern, an employee, a colleague, whatever analogy works for you. And if you actually follow that analogy through, prompting starts looking kind of ridiculous. You wouldn't walk up to an employee, drop one strangely optimized sentence on their desk and expect them to produce a perfect result. You'd brief them.

You might send them an email. Then forward another email that has some background in it. You might send three links. Maybe there's a screenshot.

Maybe you stop by their desk and say, "Okay, here's what I'm thinking."

Then you remember something fifteen minutes later and send another message. They ask a question. You answer it. You clarify something. They come back with a draft. You say, "Yeah, that's basically right, but this part is wrong."

That is how humans actually work together. Nobody calls it context engineering. It's a briefing. And I think that word carries a tremendous amount of useful information inside it.

I tend to describe certain words as little cognitive zip files. You say one thing and there's a whole structure packed inside it. Briefing is one of those words. Most people already understand what a briefing implies.

There is a goal...There is context...There may be source material...There may be constraints...There may be things the person needs to know before they start. There is probably some back-and-forth. And most importantly, the person receiving the briefing is expected to use their intelligence to figure out some portion of how to get the job done.

That last part is becoming increasingly important.

The Method Is Becoming the AI's Problem

Old prompting often looked like this: "Do this. Then do this. Format it this way. Think through these steps. Use this framework. Assume this role. Follow these ten instructions."

We were compensating for the model. We were building little procedural rails around it because it needed them. The more capable these systems become, the less sense that makes. Astra is a really good example of where this is going. The interesting thing about a model like Astra isn't simply that it may be better at writing or better at answering questions.

It's that the model increasingly has the ability to figure out the method.

That changes our job.

Instead of telling the system exactly how to do something, we increasingly need to be very clear about:

What do I want?

Why do I want it?

What does success look like?

What constraints actually matter?

What information does it need?

What authority am I giving it?

Then let the intelligence do some intelligent shit.

That's a very different relationship.

We're moving from: "Here is the procedure."

to: "Here is the objective."

That is delegation. And delegation requires something different from prompt engineering. It requires knowing what you actually want.

Intent Is Still the Hard Part

This doesn't mean you get to be lazy. Actually, I think the opposite is true.

You can absolutely throw some weak-ass bullshit at a modern model and get something back that looks surprisingly polished. We see it everywhere.

The writing sounds competent. The graphics look plausible. The strategy deck has charts. The website works. The research document has headings.

Everything looks like something happened. But there's no real intention behind any of it.

That's AI slop.

And as the models get more capable, slop may actually become easier to produce, not harder.

Which creates this weird situation where the most powerful knowledge tools humans have ever had can also become incredibly sophisticated slop machines. Pull the lever. Something impressive comes out. Pull it again.

Different impressive garbage.

It's basically a slot machine where every combination looks like you won until somebody who knows what they're looking at examines the result. And I think that is going to become increasingly unacceptable.

People are already learning to recognize AI slop. Organizations are going to get better at recognizing it. Customers will. Employers will. Audiences will. The novelty of "holy shit, AI made this" is disappearing.

The bar is becoming: Did this accomplish something useful?

That comes back to intention. Before you think about ChatGPT, Astra, Claude, agents, prompts, workflows, context windows, harnesses or any of this other stuff, you need to be able to answer a much older question: What the hell are you actually trying to do?

That has nothing to do with AI. That is a thinking problem. And AI can absolutely help you work through it. You can tell the system, "I know roughly what I'm trying to accomplish, but I haven't crystallized it yet.

Help me think through the objective." That's perfectly legitimate.

In fact, that may be one of the better uses of these systems. Talk it through. Argue with it. Change your mind. Ask it what you're missing.

Have it show you alternatives. Use the intelligence to sharpen your own intention before you pull the lever. But eventually, there has to be a there there.

You need a goal.

Otherwise you just have a very expensive machine capable of producing extremely convincing motion.

Prompt Engineering Was a Transitional Skill

I don't think prompt engineering was bullshit. Quite the opposite. It was necessary.

It taught us how models behaved.

It taught us that context matters.

It taught us that examples matter.

It taught us that ambiguity matters.

It taught us that structure matters.

It forced people to become more precise about what they wanted from a machine.

Those were useful lessons. But the machine has changed. And when the machine changes, the interface changes.

We went from: prompt engineering

to: context engineering

to: intent engineering

and I think we can finally just simplify the whole damn thing.

Brief it.

Give it what you would give a smart employee.

Here's what we're doing.

Here's why.

Here are the files.

Here are the screenshots.

Here are the links.

Here's what I'm worried about.

Here's what I've already tried.

Here are the things you absolutely cannot do.

Here's the result I need.

Tell me what I'm missing.

Then go.

That's understandable. That's teachable. And, more importantly, that is much closer to the way these systems are beginning to work.

If we really are instantiating something that behaves more and more like useful intelligence, then eventually we have to stop treating it like a vending machine that dispenses paragraphs when we enter the correct magic words.

You have to intelligently deal with the intelligence. That doesn't mean anthropomorphizing it. It doesn't mean pretending it's a person. It means recognizing that once the system can reason across context, choose methods, operate tools, revise its approach and pursue an objective, the interaction model starts looking much more like management than programming. And management has always depended on the quality of the brief.

The Good News Is This Gets Easier

There's another reason I like the word briefing. It makes AI less intimidating. We've spent years surrounding this technology with terminology that accidentally tells normal people they're unqualified to use it. You don't need to become a prompt engineer. You don't need to study some collection of magic templates. You don't need to learn how to make ChatGPT pretend it is "a world-class marketing strategist with 25 years of experience."

Just explain what you're doing. Give it the relevant material.

Tell it what you want. Answer its questions. Correct it when it misunderstands you. Keep going until the objective is clear.

That's work. That's collaboration. That's a briefing.

And as these increasingly capable models make their way into the hands of people paying twenty bucks a month instead of only enterprises spending millions, that distinction becomes enormously important. Because the capability is going to be everywhere.

The differentiator won't simply be who has access to the AI. It will be who knows what they want to do with it. That is the skill. Not prompting.

Not some clever sequence of words. Not a giant library of incantations.

Clarity of intent.

Know what you want.

Gather the context.

Explain the objective.

Brief the intelligence.

Then let it work. And if you don't know what you want yet?

Fine. Brief it on that problem first.

---

Rich Washburn is a technologist, strategist, and Founder & Chief AI Architect of ARIA AI Labs, working at the intersection of AI, infrastructure, communications, and capital. He also serves as Managing Partner and Chief AI Officer at Eliakim Capital.