A prompt is a brief
If you've ever written a creative brief, a customer-service call script, a contract clause, an SOP, or a casting note — you've already written prompts. You just didn't know they were called that, and you didn't know the audience would turn out to be a probability distribution over tokens.
You already know how to write a brief. That's 80% of prompting.
This lesson exists because most "intro to prompting" content treats the skill as new. It isn't new. The deliverable changed; the craft didn't. The craft is the same one that produced every good ad campaign, every clear SOP, every cleanly-drafted contract, every well-run support team. You name a role. You name a task. You give the context. You specify the shape of the output. You provide examples. Then you ship the brief and the work comes back.
What changed is who's reading the brief and how literally they interpret it. The model reads it like a junior contractor on their first day: literal, fast, eager, with no shared history and no ability to ask clarifying questions. That's it. That's the whole adjustment.
Side by side: a 1995 creative brief and a 2026 LLM prompt
Here is a real-shape creative brief from a 1995 agency. Read it.
Client: Midwest Furniture Co-op Project: Spring catalog cover Audience: 35-55, suburban, household income $60-120k, owns home Mood: Warm, traditional, "weekend morning" Constraints: Must show three pieces from the new oak line. No children in shot. No fluorescent light. Logo bottom-right. Reference: Pottery Barn 1994 fall catalog, pages 12-15 Deliverable: One 8x10 transparency, suitable for offset print Deadline: Tuesday
Now here is a 2026 LLM prompt for the equivalent digital task. Read it next to the brief above.
You are a senior catalog art director. Generate a description of a spring catalog cover photograph for a Midwest furniture co-op. Audience: 35-55 suburban homeowners, household income $60-120k. Mood: warm, traditional, "weekend morning." Must feature three pieces from the new oak line. No children in shot. No fluorescent light. Logo bottom-right. Reference: Pottery Barn 1994 fall catalog, pages 12-15. Output: a single paragraph, 80-100 words, suitable for handoff to a photographer.
They are the same brief. The 2026 version has the role written explicitly ("you are a senior catalog art director") and the output shape stated in tokens ("a single paragraph, 80-100 words"), but every other field maps one-to-one. Audience, mood, constraints, references, deliverable. Same.
If you wrote the first one in 1995, you can write the second one in 2026. The skill is intact. The skill is portable.
Two modes of prompting: user vs. builder
There is a distinction in this course you need to get straight early, because it will come back in every later chapter:
-
Prompts as artifacts (user mode). You sit at the ChatGPT or Claude window, you type a request, the model responds, you read the response, you adjust the request, you go again. The prompt is a thing you write once and throw away. This is what most people mean when they say "prompting."
-
Prompts as system components (builder mode). The prompt lives inside a program. The program runs it a thousand times a day, against a thousand different inputs, and ships the output to a customer who never sees the prompt. The prompt is a piece of infrastructure. It has tests, version control, and a change log. This is what the rest of this course is about.
The 5 knobs you'll learn next apply to both modes. The practice is identical. What changes is the cost of getting the prompt wrong. A bad user-mode prompt wastes thirty seconds of your time. A bad builder-mode prompt that ships into production wastes an Air Canada refund lawsuit, a NYC government scandal, a swearing-bot PR cycle. (We'll cover all three in ch24. They each started with a prompt nobody treated as infrastructure.)
For ch00, you're going to learn the discipline in user mode, because that's the mode you can practice today without writing any code. By ch19, you'll be writing builder-mode prompts. By ch24, you'll be testing them. By ch26, you'll be deploying them into agent harnesses that run them at scale.
But it all starts here: a prompt is a brief. You already know how to write a brief. Let's name the parts.
A prompt is a brief
If you've ever written a creative brief, a customer-service call script, a contract clause, an SOP, or a casting note — you've already written prompts. You just didn't know they were called that, and you didn't know the audience would turn out to be a probability distribution over tokens.
You already know how to write a brief. That's 80% of prompting.
This lesson exists because most "intro to prompting" content treats the skill as new. It isn't new. The deliverable changed; the craft didn't. The craft is the same one that produced every good ad campaign, every clear SOP, every cleanly-drafted contract, every well-run support team. You name a role. You name a task. You give the context. You specify the shape of the output. You provide examples. Then you ship the brief and the work comes back.
What changed is who's reading the brief and how literally they interpret it. The model reads it like a junior contractor on their first day: literal, fast, eager, with no shared history and no ability to ask clarifying questions. That's it. That's the whole adjustment.
Side by side: a 1995 creative brief and a 2026 LLM prompt
Here is a real-shape creative brief from a 1995 agency. Read it.
Client: Midwest Furniture Co-op Project: Spring catalog cover Audience: 35-55, suburban, household income $60-120k, owns home Mood: Warm, traditional, "weekend morning" Constraints: Must show three pieces from the new oak line. No children in shot. No fluorescent light. Logo bottom-right. Reference: Pottery Barn 1994 fall catalog, pages 12-15 Deliverable: One 8x10 transparency, suitable for offset print Deadline: Tuesday
Now here is a 2026 LLM prompt for the equivalent digital task. Read it next to the brief above.
You are a senior catalog art director. Generate a description of a spring catalog cover photograph for a Midwest furniture co-op. Audience: 35-55 suburban homeowners, household income $60-120k. Mood: warm, traditional, "weekend morning." Must feature three pieces from the new oak line. No children in shot. No fluorescent light. Logo bottom-right. Reference: Pottery Barn 1994 fall catalog, pages 12-15. Output: a single paragraph, 80-100 words, suitable for handoff to a photographer.
They are the same brief. The 2026 version has the role written explicitly ("you are a senior catalog art director") and the output shape stated in tokens ("a single paragraph, 80-100 words"), but every other field maps one-to-one. Audience, mood, constraints, references, deliverable. Same.
If you wrote the first one in 1995, you can write the second one in 2026. The skill is intact. The skill is portable.
Two modes of prompting: user vs. builder
There is a distinction in this course you need to get straight early, because it will come back in every later chapter:
-
Prompts as artifacts (user mode). You sit at the ChatGPT or Claude window, you type a request, the model responds, you read the response, you adjust the request, you go again. The prompt is a thing you write once and throw away. This is what most people mean when they say "prompting."
-
Prompts as system components (builder mode). The prompt lives inside a program. The program runs it a thousand times a day, against a thousand different inputs, and ships the output to a customer who never sees the prompt. The prompt is a piece of infrastructure. It has tests, version control, and a change log. This is what the rest of this course is about.
The 5 knobs you'll learn next apply to both modes. The practice is identical. What changes is the cost of getting the prompt wrong. A bad user-mode prompt wastes thirty seconds of your time. A bad builder-mode prompt that ships into production wastes an Air Canada refund lawsuit, a NYC government scandal, a swearing-bot PR cycle. (We'll cover all three in ch24. They each started with a prompt nobody treated as infrastructure.)
For ch00, you're going to learn the discipline in user mode, because that's the mode you can practice today without writing any code. By ch19, you'll be writing builder-mode prompts. By ch24, you'll be testing them. By ch26, you'll be deploying them into agent harnesses that run them at scale.
But it all starts here: a prompt is a brief. You already know how to write a brief. Let's name the parts.