Learn to Talk and Think AI First
You are not learning to code. You are learning to communicate.
Noelle Russell, founder of the AI Leadership Institute, opens the foundation of this whole program. She has led AI work at NPR, Microsoft, IBM, AWS, and Amazon Alexa, and the first thing she does is take the pressure off the technology. The model is not where your difficulty lives. Your difficulty lives in the clarity of thought you bring to it.
Here is the gap she wants to close. Most people meet AI as a vending machine. You walk up, type "write me an email," and expect something finished to drop out. What you get back is technically usable, a kind of minimum viable answer, but it is generic. It has no idea who you are, who you are writing to, or what good sounds like in your world. It reads like it was written for everyone, which is the same as saying it was written for no one.
The shift this lesson teaches is from consuming AI to producing with it. Not a cleverer tool. A clearer human. When you give the same model your role, your audience, an example of what you mean, and your real constraints, the output changes completely. That practice has a name, context engineering, and by the end of this lesson you will have built your first context engineered prompt and you will understand exactly why it works.
Listen to an audio summary of this session here.
Two things, in order. First, the human clarity underneath any good prompt: your why, and the experience you actually want AI to amplify. Then a repeatable structure, the RTEC framework, that hands that clarity to AI cleanly every single time. The structure is only as good as the thinking you bring to it, which is why we start with you.
Typing short prompts, taking the first answer, quietly concluding that AI does not really understand what they need.
Briefing AI with enough of your world that it returns work which is specific, useful, and unmistakably yours.
Design AI around your life, not your life around AI.
Before a single prompt, Noelle turns the attention back to you, and she means it literally. The quality of your AI grows in step with your own self-awareness. The clearer you are about your why, your values, and the outcome you actually want, the more precisely you can steer a model toward it. Your personal development and your AI development are not separate projects. They are the same one.
This is why she opens with story rather than software. Her own drive traces back to her son and the life she wants to build around her family. That is her compass, and it points every technical decision she makes. The lesson asks you to find your own version of that. Not because it is a nice warm-up, but because a model can only amplify a direction you have actually chosen. Point it at a blur and it returns a blur, faster.
So the real design question comes first, and it is a human one. We are building AI around our lives, our businesses, and the experiences we want to create, not bending ourselves to fit the tool. You decide who you want to be and what you want to be freed from. Then you build AI to carry the rest. The machine is the amplifier. You are the signal.
Three questions that locate the work worth amplifying
To find that signal, Noelle borrows a familiar overlap. Sit with each circle honestly, then look at where all three meet. That center is not a job title. It is the specific human contribution that is yours to make, and it is exactly where pointing AI pays off most. Aim AI at the overlap and it amplifies your real edge. Aim it outside the overlap and you automate work that was never yours to do well.
The third circle is the one people guess at. To pressure test it, Noelle suggests a tool like Answer the Public, which surfaces the real questions people are typing about your topic, so "what the world needs" is grounded in evidence rather than assumption.
You do not send this anywhere. Write plainly and honestly. Everything you build later, every strong prompt, draws on the clarity you find right here. Vague here means vague everywhere downstream.
Held in your browser only. Nothing is saved or sent. Come back and edit it as it sharpens.
Every interaction sits inside four nested layers.
This model is shared across the major labs, from Microsoft and Google to Anthropic and Nvidia, as a way to build AI responsibly. Picture it as four rings, one inside the next. The outermost ring is where you live, and that is deliberate. The human experience is the first and most important layer, because everything underneath it exists to serve the decisions a person makes at the top.
Knowing which ring you are standing in changes how you think about a disappointing result. When an answer comes back weak, the instinct is to blame the model, ring three, or the infrastructure, ring four. Almost always, the fix is in ring one: the conversation you had. This lesson works the outer two rings, the layers you influence directly, and leaves the inner two for later sessions. Tap a ring to open it.
Every prompt and conversation, plus the context you bake in before a conversation even starts. This is your whole lever in this program, and it is enough to change almost everything about your results.
Which model is doing the work, and the compute, security, and privacy underneath it. Real and important, covered in dedicated sessions further into the program. Not your focus today.
What matters is not how you phrase it. It is how much of your world you hand over.
Here is the same task, written two ways. Read them side by side and notice that the second one is not more clever. It is not better wordsmithing. It simply carries the context the first one left out: a role, an audience, a tone, a boundary. That is the entire difference between an answer that fits everyone and an answer that fits the person you meant.
A serviceable, forgettable email. The model had to invent the audience, guess the tone, assume the product, and pick a format. It filled every blank with an average. Viable, yes. Yours, no.
- No role to speak from
- No audience to write to
- No example of what good sounds like
- No constraints to hold the line
An email with a clear voice, aimed at a real person, inside real limits. Often it asks a sharpening question first. The result is specific, useful, and on brand, because the model is working from your world instead of guessing at it.
- Role: conversion copywriter
- Audience: women over forty, past clients
- Example: a tone to match
- Constraints: length, plus ask first
Context engineering asks more of you upfront. You do the thinking the vending machine let you skip. That is the cost, and it is real. The return is not small: the same model, given your role, audience, examples, and limits, produces something far closer to what you actually meant, every time. AI is not the problem. Vague context is the problem.
Train your eye for what a weak prompt is missing.
This is the skill in miniature. Below is a real, well-meaning prompt that still comes back generic. Four things are absent. Tap each one you think is missing, and see whether your instinct matches what would actually fix it. Tap the gaps you spot.
Tap everything you think is missing. Two picks reveal the answer.
This prompt feels reasonable, which is exactly why it is a useful trap. It names a topic and nothing else. There is no role, so the model writes as a generic everyone. No single task, just "some ideas," so it sprawls. No example, so it cannot match your voice. And no constraints, so it has no idea who the posts are for, how many you want, or what tone fits. Add those four and the same request becomes a brief the model can actually act on. That is RTEC, which is next.
RTEC: the four parts every strong brief carries.
RTEC is Role, Task, Examples, Constraints. It is not a rigid form you fill out every time. It is a checklist you run in your head before you send, the same four things you would naturally give a capable new colleague before handing them real work. Miss one and the model fills the gap with an average. Supply all four and you have written a brief, not a wish.
Take them one at a time. Each card below opens with what the part is, why it matters, and a small move that makes it stronger. Tap each letter.
Tap each letter to read what it asks of you.
One brief, assembled part by part.
Here is a complete RTEC prompt for a real task, defining a landing page headline, laid out so you can see each part doing its job. Read it top to bottom and notice how little of it is clever phrasing. Almost all of it is just context, placed where the model can use it.
You are an expert content marketer who writes warm, benefit-led copy for health and wellness coaches.
Write one landing page headline for my twelve week Mindful Eating Mastery program.
Match the feel of these two headlines I admire: "Transform Your Relationship With Food," and "Freedom From Diet Culture Starts Here." Same warmth, your own words.
Keep it under fifteen words. Empowering, never clinical. Avoid the words diet, calories, and restriction. The audience is women aged thirty to fifty-five who are tired of yo-yo dieting. Ask me any clarifying questions before you begin.
Four parts, stacked. Each is plain language a person could follow. Together they turn "write me a headline" into a brief a real marketer could act on.
You are an expert content marketer who writes warm, benefit-led copy for health and wellness coaches. Your task is to write one landing page headline for my twelve week Mindful Eating Mastery program. Here is an example of the feel I want, match the warmth, not the exact words: "Transform Your Relationship With Food," and "Freedom From Diet Culture Starts Here." Constraints: keep it under fifteen words. Empowering, never clinical. Avoid the words diet, calories, and restriction. The audience is women aged thirty to fifty-five who are tired of yo-yo dieting. Ask me any clarifying questions before you begin.Copied
Draft your first reusable prompt.
Pick one real task from your own work, something a strong brief would genuinely improve. Fill in each part and watch the prompt assemble underneath, ending with the clarifying-questions line that turns AI into a partner. This is your homework, started early. Write in your own words, then copy it into Claude.
Fill in the parts and your reusable prompt assembles underneath.
The first answer is a draft. The conversation is the product.
A strong RTEC prompt gets you a strong first response. It almost never gets you the finished thing, and it is not supposed to. The real craft is staying in the conversation and steering it. Three moves do most of that work. They feel small, but together they are the difference between using AI and thinking with it.
Read each one, then notice the pattern underneath: you are keeping the context you already built and changing only the direction. You never start over. Tap each one to see it in action.
One thread, not four. Each move keeps the context you already built and turns it forward, which is why the conversation compounds toward something yours instead of resetting.
Tap any row to expand it.
The output is close, but the tone is slightly off, or it ran long. You do not rewrite the prompt. You say, "That is good, but make the tone more conversational and cut it by about a third." The model keeps everything it already knows about your task and adjusts only the part you named. Think of it as nudging a colleague who is mostly right, not retraining them from scratch.
The first version works and you want range before you commit. Ask for more inside the same context: "Now give me three more versions of the opening line," or "Explore this from three different angles." Because the model still holds your role, task, and constraints, every new option arrives already on brief. You are widening the field of good choices, not gambling on a fresh one.
You sense the model is filling gaps with assumptions you never gave it. So you hand it the questions instead of the answers: "Ask me any clarifying questions before you begin." This is Socratic prompting. It surfaces the things you knew but had not said, and it quietly changes the relationship. When AI starts asking you questions, you have stopped being a user and become a thinking partner. If the model asks, it is because it genuinely does not know, and your answer is what makes the result yours.
Teach it once,
not every time
There is a layer above the single prompt, and it is where AI stops being a tool you operate and starts being one that already knows you. Instead of repeating your standards in every conversation, you store them once, in the model's memory or its custom instructions, and they shape everything it does by default. This is meta-prompting. It is layer two of the safety system you saw earlier, made practical.
Noelle takes it one step further into what the labs call constitutional AI. You write a short constitution for your AI: your core values and the principles you want every output to honor. She encodes her own, the Lamplighter Effect, so that everything her AI produces carries her ethics and her standard without her having to restate them. Once those principles live in memory, every task arrives already aligned with who you are.
The sequence matters. Master the full RTEC brief first, by hand, until it is second nature. Then, once you have trained a model on your context and stored it in a project or its memory, your everyday prompts can shrink, because the heavy lifting already lives in the system. "Based on this project, draft a LinkedIn post" works only because the upfront discipline was real. The shorthand is earned, never skipped.
Set once in the model's settings. Your tone, your role defaults, your do-not-do list. Every new chat starts already knowing them.
Bundles memory, files, and instructions for a whole area of your work, so every conversation inside it opens with your full context loaded.
Build your first context engineered prompt.
One thing to do before the next session, and it is built to make the lesson land in your hands rather than your notes. Pick a single real task. Write it the vending machine way first. Then write it the RTEC way. Run both. Feel the gap between them. The RTEC version is your artifact, and submitting it is how the learning locks in.
Work it in order. Each step below is small on purpose, so none of it can overwhelm. Tick them off as you go.
Tick each step as you finish it.
The reusable template, in full.
Here is the whole thing on one card, nothing to go and find. Copy it, fill the brackets with your own task, and it becomes your starting point for almost anything. Screenshot it if that helps it stick.
You are a [ROLE] helping [WHO] with [CONTEXT]. Your task is to [SPECIFIC TASK]. Here is an example of what a good output looks like: [YOUR EXAMPLE]. Constraints: [WHAT TO AVOID OR FOLLOW]. Ask me any clarifying questions before you begin.Copied
- Your role. Who is the AI in this conversation?
- Your task. What exactly do you need?
- Your example. One paragraph showing what good looks like.
- Your constraint. What to avoid or follow.
What this lesson points to, and where to go next.
Everything you need to act is here, with a direct link on each one. You do not need every tool today; reach for each as it becomes relevant.
The AI you will use for the assignment and most of the program. Start from the chat tab, paste your RTEC prompt, and iterate from there.
claude.ai ↗Speak your context instead of typing it. Voice carries detail faster and works in any input field, which lowers the friction of starting.
wisprflow.ai ↗Surfaces the real questions people search around a topic, so "what the world needs" is grounded in evidence rather than a guess.
answerthepublic.com ↗The Holistic Evaluation of Language Models from Stanford. Public leaderboards grading models on accuracy, bias, and safety, for when model choice matters later.
crfm.stanford.edu/helm ↗Sign in to the People app with your Mindvalley account, open this lesson, find the context engineering challenge, and submit your RTEC output there. That submission is what marks the lesson complete.
people.mindvalley.com ↗Noelle Russell and the AI Leadership Institute, with a free Responsible AI Bot Builder course and a copy of this presentation. The Mindvalley community space is also where you tag Almerissa for admin and Rui for AI questions.
noelle.ai ↗The terms, in one place.
You are not learning to code. You are learning to communicate.