I’ve reviewed over a thousand pitches for AI companies. 99% of them get it wrong. And it all starts with how we're teaching AI.
As an investor in companies like Anthropic, OpenAI, and Scale AI, I see where the puck is going. As a founder who sold two companies, I know what it takes to build a team that can actually execute. The disconnect between what the industry needs and what education provides has never been wider.
Most guides on AI curriculum are written by technologists who think teaching is just a matter of listing the hottest new tools. They’re wrong. I’ve seen the results firsthand: graduates who can write a Python script to call an API but have zero intuition for how these systems actually think, break, or—most importantly—create real value.
Forget everything you’ve read. This is my unconventional, classroom-tested guide for educators who want to prepare students for the real world, not just for a certification exam.
The Problem: We're Training Tool Operators, Not Thinkers
Back when we were building RemoteTeam, we needed to hire engineers to build out some early AI-powered features. The résumés were impressive, full of projects using the latest frameworks. But in the interviews, I’d ask a simple question: “Okay, you used this model. What are its potential failure points? How would you know if the training data was biased?”
Crickets.
They could operate the tool, but they couldn’t think about the system. They had learned the what, but not the why or the how. This is the fundamental flaw in modern AI education. We're so obsessed with the fast-changing landscape of tools that we've forgotten to teach the foundational principles that never change.
The hard truth is that the hot AI tool you teach today will be a relic in 18 months. The core concepts, however, are timeless.
Counterintuitive Rule #1: Stop Teaching Tools, Start Teaching Mental Models
Your goal as an educator shouldn't be to make someone a master of a specific library. Your goal should be to build a robust set of mental models they can apply to any tool, present or future.
When I talk to my portfolio founders, the biggest challenge they face isn't a shortage of people who can use an API. It's a shortage of people who can reason from first principles. So, what does that look like in a curriculum?
Instead of a module on a specific vector database, teach the concept of data representation. What does it mean to turn words, images, or sounds into numbers? What is lost in translation? Explore different embedding techniques and, more importantly, their trade-offs.
Instead of a unit on a particular model, teach systems thinking. An AI product isn't just a model; it's a complex system of data pipelines, user feedback loops, ethical guardrails, and infrastructure. Have students map out this entire system. Where is the data coming from? Where do the predictions go? Who is impacted?
Here are the core concepts I believe every AI curriculum should be built on:
- Data as the Bedrock: Not just cleaning data, but understanding its origin, its inherent biases, and the ethics of its collection. The model is just a reflection of the data it was fed. Garbage in, garbage out is the iron law.
- Probabilistic Thinking: AI is not deterministic. It's a world of probabilities. Students need a deep intuition for uncertainty. An AI that is 95% confident is still 5% wrong. What are the consequences of that 5%?
- The Cost of Inference: Every AI query has a cost—in compute, in dollars, in energy. Students should be able to do the back-of-the-napkin math. Is this feature economically viable? Does it scale?
- Ethics and Safety: This cannot be a single lecture at the end of the semester. It must be woven into every single topic. For every concept you teach, ask the question: “How could this be used to cause harm?”
Counterintuitive Rule #2: Make Them Build, But Make It Hurt
Project-based learning is essential, but most AI projects are just glorified tutorials. They follow a happy path where everything works. That’s not how the real world operates.
Real learning happens when things break. Your curriculum needs to force students off the happy path and into the mud.
At one of my startups, MovieLaLa, we were building a recommendation engine. The first version was terrible. It kept recommending the same popular movies to everyone. The team had to go back to the drawing board, not because the algorithm was wrong, but because our assumptions about the user were wrong. The pain of that failure taught us more than any textbook ever could.
How do you replicate this in the classroom?
- Give them messy data. Not the clean, pre-packaged datasets from a tutorial. Give them real-world data with missing fields, weird formatting, and human errors. Make them feel the pain of data cleaning.
- Create adversarial challenges. Once a student builds a project, have other students try to break it. Can they find inputs that make the model produce biased, incorrect, or offensive outputs? This teaches defensive thinking and robustness.
- Enforce constraints. Don't let them use a massive, pre-trained model for a simple task. Give them a tiny compute budget. Force them to use a smaller, more efficient model. Constraints breed creativity and a deeper understanding of trade-offs.
Out of the 200+ companies I've invested in, the ones that truly break through are the ones that survive this early struggle. They are forged in the fire of broken models and unhappy users. Your classroom should be a simulator for that fire.
Counterintuitive Rule #3: The Most Important Skill is Asking Questions
We are rapidly moving past the need for people who can code. The value is shifting to people who can ask the right questions.
An AI is a powerful engine, but it needs a driver. The quality of the output is a direct function of the quality of the prompt. Prompt engineering is a start, but it’s deeper than that. It’s about having the domain expertise and critical thinking skills to interrogate the model.
I once saw a team spend three months building a complex model to predict customer churn. The model was technically brilliant, a real work of art. But they never stopped to ask the sales team what the early warning signs of churn actually were. A ten-minute conversation would have saved them three months of work. They built a solution to a problem they didn’t understand.
Your curriculum must relentlessly focus on this.
- The Art of the Prompt: Go beyond basic prompt engineering. Teach students how to provide context, examples, and constraints. Teach them how to iterate on a prompt to refine the output.
- Domain-Specific Inquiry: Pair students with experts from other fields—historians, biologists, artists. Have them work together to solve a problem using AI. The AI student brings the technical skill, but the domain expert brings the critical questions. This is the future of work.
- Deconstructing Outputs: Never accept an AI's output at face value. Teach students to be professional skeptics. Where did this answer come from? What are the potential sources of error? Can I verify this with an independent source? Make them fact-check the AI.
A Sample Counterintuitive AI Curriculum Outline
So what does this look like in practice? Here’s a high-level outline that flips the traditional model on its head.
Module 1: The System, Not the Model
- Week 1: The Anatomy of an AI Product. We dissect a real-world AI product (e.g., Spotify’s Discover Weekly). We map the entire system: data inputs, user feedback loops, business KPIs, ethical risks. No code, just whiteboards and critical thinking.
- Week 2: Data’s Dirty Secrets. Students are given a horribly messy dataset from a real (anonymized) business. Their only task: document its flaws. They present their findings on data bias, missing values, and ethical landmines.
- Week 3: The Cost of a Query. We break down the unit economics of an AI feature. Students calculate the cloud computing costs, latency, and energy consumption. They must then argue for or against the feature’s financial viability.
Module 2: Building with Broken Bricks
- Week 4-6: The Constrained Challenge. Students are put into teams. They are given a real-world problem but with severe constraints: a tiny compute budget, a deliberately flawed open-source model, and a 24-hour deadline. The goal is not a perfect solution, but a clever one.
- Week 7: Red Team, Blue Team. Each team’s project is handed over to another team (the “red team”) whose sole job is to break it. They document every failure, every biased output, every security hole. The original team (the “blue team”) must then present a plan to fix the issues.
Module 3: The Art of the Question
- Week 8: The Expert Interview. We bring in professionals from non-technical fields (e.g., a journalist, a doctor). Students must interview them to understand a core problem in their field and then frame that problem in a way an AI could help solve. The deliverable is not code, but a well-defined problem statement and a series of critical questions.
- Week 9-10: The Socratic Loop. Students work on a final project, but with a twist. At every stage, they must defend their decisions to a panel of instructors and peers. Why this data? Why this model? Why this user interface? The focus is on the quality of their reasoning, not just the final output.
The Educator's Role in the Age of AI
Your role is changing. You are no longer the sage on the stage who dispenses knowledge. You are the coach, the sparring partner, the professional skeptic. Your job is to create the conditions for painful, memorable learning. Your job is to ask the questions that students aren't asking themselves.
I didn’t get to where I am by following a tutorial. I got here by making a thousand mistakes, by seeing my assumptions blow up in my face, and by learning to ask better questions. My two exits weren't because I had all the answers, but because I built teams that were relentless in their search for the right questions.
Stop training tool operators. Start forging critical thinkers. The future of AI, and the future of our economy, depends on it. Don't just teach them how to use AI; teach them how to think with it, against it, and beyond it. That’s a curriculum that matters.
Frequently Asked Questions
What tools do I need to get started?
Start with the basics. You don't need expensive software or fancy tools. A spreadsheet, a note-taking app, and direct access to your customers will get you further than any enterprise platform. Add tools only when you hit a specific bottleneck.
How long does it take to design an ai curriculum (the counterintuitive guide for educators)?
The timeline varies depending on your starting point and resources. For most founders, expect 2-4 weeks for initial setup and 2-3 months to see meaningful results. I've seen teams move faster when they focus on one thing at a time rather than trying to do everything at once.
Do I need technical skills to design an ai curriculum (the counterintuitive guide for educators)?
Not necessarily. While technical understanding helps, the most important skills are clear thinking and the ability to break problems into smaller pieces. Many successful founders I've invested in started with zero technical background and either learned enough to be dangerous or found the right technical partner.