📖 Living glossary (read first — come back whenever you need to)
This learning path is about human skills that remain yours alone. Here are the NEW product terms we’ll use — learn these before continuing:
💡 AI is bad at original ideas
🧠 Imagine it this way: the AI is a phenomenal studio musician — it plays any sheet music you put in front of it, in the right key, without a mistake. But ask it to compose the new music and something that already exists comes out, remixed. The original melody still comes from you.
There's a dangerous myth going around: "AI will come up with the ideas; I just press the button." Matt Pocock cuts right through that — AI is notoriously bad at original ideas. It makes sense for what it is: a language model learns to predict what is more likely come next, based on everything that has already been written. In other words, it’s a machine for average of the past. By design, it steers toward the conventional, toward what already exists—not toward the unprecedented leap that defines a new product.
Where the AI in fact is an advantage, according to Pocock, is in the execution — not in having the right idea. It implements quickly, writes the code, builds the screen. But deciding which product is worth it, which the problem matters, what angle is different — that remains human work. That’s why he sums it up: "you need to have the VISION and know why you're building". O common mistake is asking AI to "come up with a successful app" and accepting the first cliché it returns: you just outsourced the very part that was your advantage. AI is the execution engine; the direction is yours.
AI gravitates toward what’s likely (the center); what’s new emerges outside the cloud—with you.
⚠️ Common beginner mistake
Treat the AI like a "startup founder": ask for the idea, the name, the audience, the features — all at once. You get a generic jumble and lose the one thing that set you apart: your vision. Use AI to execute your idea, not to come up with ideas for you.
In one sentence: the AI executes your idea quickly, but the original idea still has to be yours.
Going deeper (optional): why does an LLM converge on the conventional?
A language model is trained to predict the most likely next piece of text given everything that came before. “Most likely” = most frequent in the training data = most conventional. That’s why, without strong direction from you, its default output tends toward the commonplace. You can push it beyond that with a highly specific, opinionated prompt — but the opinion (the vision) has to come from you.
🎯 You choose the features
🧠 Imagine it this way: a chef who accepts EVERY ingredient that shows up in the kitchen ends up with a confusing dish. A good chef chooses a few of the right ingredients and discards the rest. Each feature is an ingredient — and you're the chef, not the stock clerk.
Pocock’s point is straightforward: "YOU choose the features". It seems obvious, but with AI the temptation takes a different shape. Before, adding a feature used to take days of work, so you thought twice. Now AI builds any feature in minutes—and then the instinct becomes "if it's cheap, add everything." That's exactly the trap. The cost of a feature was never just writing it: it's maintaining it, testing it, and, above all, the confusion that it adds for the user.
Pocock gives a concrete counterexample: don't become "a VC-funded app with a thousand features". Companies flush with investor money (VC-funded) tend to stack feature on top of feature until no one understands what the product is for anymore. With AI making it cheaper to build, anyone can fall into this trap without product vision. Choosing a feature is really choosing what no do. AI is willing to build everything; it’s up to you to say "these three, and only these."
Quick recall: how does cheap AI change the game for features?
In one sentence: with AI, deciding what NOT to build has become the most valuable work—and it’s yours.
✂️ Ask what to remove
🧠 Imagine it this way: A sculptor doesn’t ask, “What stone should I add?” They remove stone until the statue appears. The best product is like that — what you remove reveals the shape. Point the AI’s chisel at this.
Here's the most practical and counterintuitive advice in the module. Most people use AI like this: "what can I add in my app?". Pocock turns the question on its head: "You should be asking AI what to REMOVE from your app, how to make it simpler, how to improve UX." In other words—ask the AI what remove, how simplify, how to improve the UX. AI is great in this role because it can see the whole codebase at once and point out overlaps, useless screens, and awkward flows you no longer notice.
Notice the role reversal. Adding features is where AI would be dangerous (it makes things up generically; remember topic 1). But remove and simplify is where it becomes your ally — because you still make the final cut, and it only gives you the candidates. You keep the vision; it does the cleanup. The common mistake is just growing the app without ever pruning it: it becomes that software that does everything and does nothing well. Make "ask what to remove" a recurring habit in your workflow.
🔬 Worked example: the notes app that kept growing
You built an AI notes app. In two weeks, it had notes, tags, folders, reminders, kanban, a productivity chart, AI chat, a customizable theme, and export to 5 formats. No one could make sense of the home screen anymore. Instead of asking "add advanced search," you flipped it around:
The prompt
"Look at my notes app. List 5 features I should REMOVE to simplify it, ranked by how much they confuse users. For each one, say what UX gains without it."
The result
AI flagged kanban, the chart, and 3 of the formats as noise. You cut 4 of the 5 (kept the chart; it was your differentiator). A simpler, faster, clearer app—and the VISION remained yours: you made the final call on what to cut.
In one sentence: ask AI what to REMOVE, not what to add — and make the final call yourself.
📐 Product design hasn't changed
🧠 Imagine it this way: they invented the electric drill and the automatic screwdriver — but the rules for building a house that won’t collapse are still the same. The tools got faster; the architecture didn’t.
There's a narrative that "AI changed everything, throw out the books." Pocock actually disagrees: "I don't think much has changed." What changed was the speed of execution. The fundamentals of product design — understanding the user, finding the real problem, designing the simplest solution — all remain exactly the same. In fact, he says that the classic product design books still hold up. The AI hasn't revoked any of that; it just lets you test your design decisions faster.
This connects with everything you saw in this track: AI has "eaten" the part tactical (the execution), and the most valuable part left over is the strategic (the product decision). Knowing how to choose, cut, prioritize, and design the experience is a muscle you train — not a download AI does for you. The common mistake is thinking that because the code comes out ready, you can skip product thinking. It’s the opposite: now that execution is cheap, think through the product is where you win or lose.
⏩ What AI sped up
- • Write the code and build screens.
- • Make prototypes to test an idea.
- • Iterate quickly on a design decision.
🧱 What stays the same (yours)
- • Understand the user and the real problem.
- • Deciding what goes in and what comes out.
- • Design the simplest experience.
In one sentence: the AI sped up execution, but the fundamentals of product design are still the same — and so are yours.
🗣️ Talk to clients
🧠 Imagine it this way: you can have the fastest kitchen in the world, but if you never ask the customer what they want to eat, you'll cook beautiful dishes that no one orders. Talking to the customer is the menu; AI is just the fast stove.
When Pocock is asked what changed about building a business in the AI era, his answer is calm: "I don't think much has changed." The fundamentals remain solid—and the first one is talk to customers. Discovering what they in fact need, which pain is real, which problem is worth solving. No LLM has access to your actual customers; this information only makes it into the product if you seek it out in the conversation.
The classic flow still fully applies: talk to customers → understand the need → build prototypes that solve the real problem → show them back to customers. The difference is that AI makes the middle step (building the prototype) much faster — so you can run this cycle more often. But if you skip the conversation, you’ll use AI’s speed to build the wrong thing, very quickly. The common mistake is locking yourself away with AI "producing features" and never facing a customer. Speed without direction from customers just gets you to the wall faster.
In one sentence: talking to customers remains the engine of the right product—AI just speeds up the prototype.
🚀 Build an AI business
🧠 Imagine it this way: the AI is a turbo engine. Put it in a car going in the right direction, and it flies. Put it in a car pointed at a cliff, and you just reach the cliff faster. The steering wheel (vision) decides everything.
Putting it all together: building a business with AI is, fundamentally, building a business — the fundamentals haven't changed. AI gives you an edge in execution, not having the right idea. So the game is: you have the vision, talks to customers, picks a few of the right features, asks AI what remove, and uses its speed to run the product cycle more times. Those who outsource their vision to AI build generic things quickly. Those who keep their vision and use AI as an execution engine build what in fact solves the customer’s problem. Before asking AI to build anything, run the checklist below — it’s the practical summary of this module. Copy and paste it at the start of each project:
Antes de mandar a IA construir, garanta que a VISÃO é SUA: [ ] VISÃO — sei o que estou construindo e POR QUÊ (não pedi a ideia pra IA)? [ ] CLIENTE — falei com gente real e sei qual dor estou resolvendo? [ ] FEATURES — escolhi poucas, certas (não virei um app de mil features)? [ ] REMOVER — perguntei à IA o que TIRAR / simplificar / melhorar a UX? [ ] EXECUÇÃO — usei a IA só pra implementar e prototipar mais rápido? Se a IA está decidindo o produto por você, pare: o volante é seu.
Quick recall: in the AI era, where is its real advantage for a business?
In one sentence: the AI is the turbo engine; vision is the steering wheel — and the steering wheel is still in your hands.
🧾 Module Summary
Next track:
Track 3 — AI Skills (Skills): how to package your knowledge into procedures the AI always follows your way.