The app inflation

Two-thirds of the world’s apps were created in the last three or four years

What it is

The phenomenon of having too many apps, created too quickly, because AI tools brought the cost of building software close to zero.

Why learn this

When everyone can launch, launching stops being a differentiator. Understanding app inflation helps you avoid confusing “I managed to build it” with “I have a product.”

Key concepts

App inflation, marginal cost of building, excess supply, scarce attention.

An estimate often cited by people who follow the software market says that about two-thirds of the applications in existence today were created in the last three or four years. The exact number matters less than the trend: the amount of software grew much faster than the number of people willing to use it.

Terms in this topic

• App / application: Any software someone uses to complete a task: a website, phone app, extension, or web dashboard.
• SaaS: Software as a Service. You don’t buy the program; you pay a monthly subscription to use it over the internet (e.g., Notion, Canva, RD Station).
• Vibe coding: Building software by describing in everyday language what you want to an AI (Claude Code, Codex, Cursor) and letting it write the code.
• Marginal cost: How much it costs to produce one more unit. With AI-built software, the cost of “one more app” has dropped to a few hours and a few dollars.

Inflation, in the economic sense

In economics, inflation is when there’s too much money in circulation and each bill becomes worth less. Something similar happens with apps: there’s too much software competing for the same attention, and each individual launch becomes worth less to the people who see it.

That doesn’t mean software has lost its value. It means generic software has lost its value. What still matters is software that solves a real pain point in a way that’s hard to copy and reaches the right people.

What got easier

• Go from an idea to a working prototype in hours.
• Test several versions without hiring anyone.
• Build on your own what used to take a team.

What got harder

• Get noticed among thousands of similar launches.
• Convince someone to pay for one more subscription.
• Keep an advantage when anyone can copy you in days.

Practical consequence

Before opening the editor, ask the question that inflation forces you to ask: if ten more apps just like mine appear tomorrow, why would anyone still choose mine? If there’s no answer, the problem isn’t technical.

Is SaaS dead?

Where there’s still room for arbitrage in a saturated market

What it is

The debate over the “SaaS apocalypse”: the idea that software subscriptions lose their purpose when anyone can build their own tool with AI.

Why learn this

The idea is partly right. The opportunity lies in knowing where it gets things wrong.

Key concepts

SaaS apocalypse, arbitrage, niche, use case, speed.

Many people keep saying SaaS is over. They’re right in many ways: simple tools that just organized data or generated text are being replaced by a two-line request to an AI assistant. Why pay R$ 50 a month for something the chat does for free?

Terms in this topic

• Arbitrage: Making money from a difference the market hasn’t corrected yet. Buying cheap in one place and selling high in another; here, solving a problem before the rest of the market realizes it’s easy to solve.
• Niche: A specific group of customers with a specific need: accountants in small towns, veterinary clinics, engineering firms.
• Use case: The specific task someone wants to do with a product: “issue this month’s invoice,” not “manage finances.”

The three places where there’s still room

• The right niche: groups with little competition, money to spend, and little time to learn AI.
• The right use case: a narrow, painful, frequent task instead of a generic platform.
• Speed: Get there first and learn faster than others can copy you.

Common trap

Read “SaaS is dead” and give up, or read “there’s still room” and build anything. Both takes are lazy. What you face is a stricter filter: fewer ideas make it through, but those that do face less serious competition.

The product cycle got scrambled

Idea, validate, prototype, marketing, develop, launch, improve—in a different order

What it is

The classic product life cycle and how AI changed the order and cost of each step.

Why learn this

Those who stick strictly to the old order lose the advantage of building quickly. Those who ignore the order completely build without knowing whether anyone wants it.

Key concepts

Product life cycle, validation, prototype, order of steps, iteration.

The traditional product process has seven steps: come up with an idea, validate it, prototype, market it, develop it, launch it, and improve it. It was a logical sequence back when development was expensive: you only spent money on engineers after you were sure it was worth it.

With AI, building a prototype takes an afternoon. That changes the logic: sometimes it makes more sense to prototype to validate than to validate and then prototype. A working prototype answers questions an opinion survey never can, like “does the person actually click?”

How the order usually works today

1. An idea grounded in a pain point. Ideation is still at the center. Nothing replaces understanding the problem.
2. Quick prototype. An ugly but usable version, built in hours with a coding assistant.
3. Validation with the prototype in hand. You show something real to real people and measure what they do, not what they say.
4. Distribution from day one. Marketing is no longer the final step. Without an audience, there’s no way to validate anything.
5. Develop, launch, and improve in short cycles. The final steps become an ongoing cycle instead of a straight line.

Simple rule

The right order depends on the product and the niche. Ask which step is the cheapest to do right now and which answers the most dangerous question. Start with the one that does both.

When experience is already validation

The micro-SaaS for a problem you’ve solved ten times

What it is

A situation where you’ve solved the same problem many times for customers or colleagues and want to turn that solution into a standardized product.

Why learn this

It’s the safest shortcut for someone just starting out: the need has already been proven in your own work.

Key concepts

Micro-SaaS, making a process repeatable, standardizing, validation through experience, relative price.

If you’re a consultant, freelancer, or have worked in a field for years, you’ve probably noticed a recurring gap: every customer needs the same little thing, and no one offers it in a simple way. In that case, you don’t need as much validation. Your experience has already been your market research.

Terms in this topic

• Micro-SaaS: A small SaaS focused on a single task, often built and maintained by one or two people.
• Make a process repeatable: Turn something you do “by hand,” one case at a time, into a repeatable process that works the same way every time.
• Plug and play: Something a person connects and that just works, without complicated configuration.

The customer’s math

The customer compares your price with the alternative. If the alternative is hiring a developer for a few weeks, or paying for a premium AI subscription that costs hundreds of dollars a month and still requires them to learn how to use it, a R$ 49 product that solves that exact problem seems cheap.

The price isn’t compared with zero. It’s compared with the cost and work of solving the problem another way.

Signs your experience counts as validation

• You’ve done this for five or more different customers.
• Customers asked you to do it again or referred others.
• You charge for it today, even if you do it manually.

Signs you still need to validate

• You’ve only solved it for yourself.
• Every customer asked for a different version.
• No one would pay if you weren’t there.

One person, three roles

Product, engineering, and validation in one head

What it is

The possibility of one person, supported by AI agents, doing work that used to require a product manager, engineer, and quality analyst.

Why learn this

That is power and a trap at the same time: you can do everything, but you can also get everything wrong with no one to question you.

Key concepts

Product manager, engineer, QA, AI agent, self-criticism.

In companies, products are usually made by people in separate roles. The product manager decides what to build and for whom. The engineer builds it. The QA tests it and finds problems. Each person holds the others in check, and that friction prevents many mistakes.

Terms in this topic

• Product manager (PM): The person who decides what the product does, who it is for, and the order in which things are built.
• QA: Quality Assurance. The person who tests the product to find errors before the user does.
• AI agent: An AI model that carries out tasks with several steps on its own: reads files, runs commands, tests, and fixes things.

The role that tends to disappear

People who know how to program tend to jump straight to engineering because it is the fun part. AI agents make this worse: building becomes so easy that no one stops to ask whether they should.

That is why the rest of this course is almost entirely about the product role: the questions a good manager would ask before releasing a line of code.

Wear one hat at a time

Set aside different times for each role. In one session, only make decisions (product). In another, only build. In another, only try to break what you built. Mixing all three at once means the builder always wins.

Speed loves money

Why getting there first matters more than ever

What it is

The principle that in a market where anything can be copied, the window of opportunity is short, and whoever moves quickly captures most of the value.

Why learn this

Today’s advantage lasts months, not years. Planning as if it will last for years is the most common way to lose it.

Key concepts

Window of opportunity, first to arrive, short cycles, fast learning.

There is an old saying in business: speed loves money. It has always been true, but it matters more today. If any product can be cloned in weeks, the advantage of being first is measured in weeks too.

Speed that helps

• Launch the minimum version and learn from real use.
• Make decisions in days that used to take months.
• Quickly abandon what does not work.

Speed that gets in the way

• Skip the question, "Is this necessary?".
• Launch something broken that ruins the first impression.
• Change your mind every week without finishing anything.

Speed is not haste

Being fast means shortening the time between a decision and the learning it brings. Haste means skipping the thinking. The questions in the next tracks take an afternoon; they are what make speed safe.

Measure in days

For each idea, set a short deadline (seven to fourteen days) to get a first version into someone else’s hands. If it cannot fit that deadline, the idea is probably too big.

Module 1.1 summary · The game has changed

• App overload: there’s more software than people have attention for; generic products have lost value
• Is SaaS dead?: generic products, yes; niches, use cases, and speed still leave openings
• Mixed-up cycle: a low-cost prototype can come first and help validate the idea
• Experience as validation: solving the problem many times is often the safest shortcut
• Three roles: one person can handle product, engineering, and QA; product is the part that gets left out
• Speed: the window lasts months; speed doesn’t mean rushing

Building with LLMs means building the average

Why the model pulls everything toward the usual

What it is

The tendency of language models to produce the most likely answer, which is also the one most like what already exists.

Why learn this

If you let the model decide the design, text, and journey, your product will look like all the others made the same way.

Key concepts

LLM, most likely answer, statistical average, pattern, generic.

Put simply, a language model works by predicting the most likely continuation of a text. It learned this by reading a huge amount of content that already exists. The natural result is that without clear guidance, it gives you what’s most common: the most common layout, the most common marketing phrase, the most common journey.

Terms in this topic

• LLM: Large Language Model. The type of AI behind ChatGPT, Claude, and Gemini.
• Average (here): The most typical result, the one that appears most often in the training data. It’s not bad; it’s just the same as everything else.
• Slop: Slang for careless AI-generated content: correct, generic, and soulless. You can spot it from a mile away.

Average isn’t a flaw; it’s the nature of the tool.

For tasks where ordinary is the right choice (a sign-up form, a confirmation email), average is exactly what you want. The problem comes up when you use the same tool in the same way for the part of the product that should be unique.

Where to accept average

Let the model decide what counts as infrastructure (authentication, tables, settings screens). Save your judgment for three things: the main promise, the user’s first experience, and the visual identity.

Pattern interrupt

Average doesn’t make anyone stop scrolling

What it is

A stimulus that breaks the viewer’s expectations and demands attention: something different from what they’ve seen a hundred times.

Why learn this

In an endless scroll, anything that looks the same disappears. Your product needs at least one element that makes people stop.

Key concepts

Pattern interrupt, attention, expectations, differentiation, first impression.

Think about your feed. You scroll past dozens of posts a minute without reading them. What makes you stop is something that breaks the pattern: a strange image, an unexpected phrase, a product that does something you didn’t know was possible. In marketing, this is called a pattern interrupt.

Terms in this topic

• Pattern interrupt: The moment something breaks from what’s expected and captures attention by reflex.
• Endless scroll: Feeds that never end (Instagram, TikTok, X), where the content keeps coming and attention lasts for fractions of a second.

The problem with average connects here: once everyone has seen the average version, it stops interrupting. The first site with a certain purple gradient caught people’s attention. The thousandth didn’t. The same goes for features: the first app that summarized meetings impressed people; today, it’s expected.

Interrupts

• A demonstration that solves something in seconds, right in front of someone.
• An unexpected positioning for a familiar problem.
• A visual style that clearly didn’t come from a standard model.

No longer interrupts

• “With the power of AI” in the headline.
• A landing page with the same gradient, icons, and phrases as all the others.
• Another generic chat assistant.

The erosion of judgment

When everything is outsourced, even the user journey

What it is

The risk of stopping thinking for yourself because AI always offers a ready-made answer, including for product decisions.

Why learn this

Judgment is what sets apart people who build products that work. When it atrophies, you only produce the average.

Key concepts

Judgment, cognitive outsourcing, discernment, user journey, independent thinking.

We’re entering a world where few people will exercise their own judgment. Everything can be outsourced: the writing, the design, the strategy, even the user journey, the sequence of steps someone takes from arriving at your product to getting what they wanted.

Terms in this topic

• Judgment: The ability to make good decisions without having all the information: knowing what matters, what to set aside, and when to stop.
• User journey: A person's complete path through a product: discovering it, getting started, understanding it, using it for the first time, and coming back.
• Cognitive outsourcing: Delegating not only the task but also the thinking about the problem to someone else (a person or AI).

Why this becomes an advantage for those who resist

If most people outsource their judgment, those who keep using their own become rare. And rarity, in a crowded market, is valuable. Being careful, thinking about the user experience, questioning the model's suggestion: all of this becomes a competitive advantage.

One-week exercise

Before asking AI to make any product decision, write your own answer in three lines. Only then compare it with the model's. You'll notice where you agree out of laziness and where you had a better idea.

The word "launch" has lost its meaning

Novelty fatigue and favorites no one rereads

What it is

The constant flood of ads for new products that makes people stop paying attention to any launch.

Why learn this

If your promotion plan is "announce it and wait," you’re competing with dozens of ads a day, and you’ll lose.

Key concepts

Launch fatigue, "introducing," favorites, attention, noise.

Open any tech social network and count how many times the word "introducing" appears in a day. A new AI QA engineer. A faster new voice model. A new agent that does everything. There are so many that the word has lost its weight.

Typical behavior in the face of too much

• The person scrolls for 40 minutes and saves six things "to look at later."
• The next day, six more appear. Yesterday’s are never opened again.
• The favorites folder becomes a graveyard of good intentions.

What this means for your product

• Deliver the first dose of value very quickly, preferably on the first visit.
• Solve a pain point so well that the person feels immediate relief, not curiosity.
• Be remembered for a concrete reason, not for being "another launch."

The same goes for design "tips"

Many videos promise "the seven skills for a design that doesn’t look AI-made," and the result still looks AI-made. Following the trendy formula means, by definition, producing the trend’s average result.

A crowded market isn’t a closed market

There are still gaps that are poorly solved or poorly distributed

What it is

The idea that even with too many products and few moats, there are always problems left that no one has solved well or brought to the right people.

Why learn this

It helps you avoid the opposite of overconfidence: thinking it’s no longer worth building anything.

Key concepts

Gap, moat, distribution, problem precision, underserved niche.

If everything can be copied and no one has a moat, does any product deserve to exist? No. When you look at a problem with enough precision, you almost always find parts no one has solved, parts solved the wrong way, or parts that were solved but never reached the people who need them.

Terms in this topic

• Moat: What protects a business from competition: brand, exclusive data, switching cost, user network. The image is a moat filled with water around a castle.
• Distribution channel: The path the product takes to reach the customer: social media, partners, Google search, referrals, direct sales.

Questions that reveal gaps

• Who solves this today with a spreadsheet or on paper?
• Which existing solution do customers complain about using?
• Which group is left out because the tool is difficult or expensive?

Questions that hide gaps

• "Does something similar already exist?" (Almost always.)
• "Is this innovative?" (Innovation is not a requirement.)
• "Can this be done with AI?" (Almost anything can.)

Distribution is a gap, too

Often, the solution exists but has never reached the right audience. Bringing a familiar solution to an underserved niche, in their language and in a way that works for them, is a legitimate opportunity.

The three kinds of judgment

Knowing how to build, knowing what to build, knowing whether it’s worth building

What it is

The three levels of judgment people need to create products today: technical skill, problem selection, and value assessment.

Why learn this

AI has made the first level almost free. The advantage has shifted to the other two.

Key concepts

Technical judgment, problem judgment, value judgment, discernment.

Building in a time when almost anything can be built requires judgment at three levels. The first is knowing how to build. The second is knowing what to build. The third, and rarest, is knowing whether it’s worth building.

The three kinds of judgment side by side

• Building · Question: How do I make this work? · Who does this well now: Anyone with a coding assistant
• What to build · Question: Which problem should I solve? · Who does this well now: People who know a niche well
• Whether it’s worth building · Question: Does this deserve to exist a year from now? · Who does this well now: Few people. This is what this course trains you to do.

How to use all three in practice

1. Start with the third. Before any prototype, answer this: Is it worth it? The entire Track 2 is about this question.
2. Then the second. If it’s worth it, which specific part of the problem will you tackle first?
3. Only then the first. Now you can build quickly, with all the help AI offers.

A sentence to stick on your monitor

In a world where you can build anything, judgment about what’s worth building is the whole game.

Module 1.2 summary · The average problem

• The average: An LLM without direction delivers the most common result
• Pattern interruption: What everyone has already seen won’t make anyone stop
• Judgment: Delegate the execution; keep the product decisions
• Launch fatigue: Value needs to arrive before the product becomes a forgotten favorite
• Gaps: Even a crowded market has problems that are poorly solved or poorly served
• Three kinds of judgment: How to build, what to build, and whether it’s worth building; the last is rare

Productive pessimism is not defeatism

Imagine the worst so you can act better

What it is

The practice of deliberately listing the ways a plan could fail, with the goal of reducing risks, not justifying inaction.

Why learn this

Optimism without testing leads to products no one wants. Pessimism without action leads to nothing. The middle ground works.

Key concepts

Productive pessimism, defeatism, optimism bias, risk management.

People who start businesses tend to be optimistic, and that is necessary to get started. The problem is that optimism can also hide risks. Productive pessimism is the antidote: you deliberately and carefully imagine all the ways the plan could fail.

Terms in this topic

• Defeatism: Using the risk list as an excuse not to try. It’s pessimism that keeps you from moving forward.
• Optimism bias: The human tendency to think things will work out better for us than for other people.
• Pre-mortem: A technique for imagining that a project has already failed and asking: why? You do this before you start.

Productive mindset

• “If no one knows it exists, how can I get the first 10 users to find it?”
• “If someone copies it in a week, what would still be mine?”
• “If I’m wrong, how much will I lose?”

Defeatist mindset

• “I don’t have an audience, so there’s no point.”
• “They’ll copy it anyway, so why try?”
• “Something similar already exists. I give up.”

One-line test

For every risk you list, write an action next to it that takes less than a day. If there isn’t one, that risk is serious and deserves attention before anything else.

The contingency journal

“If I lost everything tomorrow, how would I get by?”

What it is

A habit of recording concrete plans over the years for surviving the worst-case scenario, such as losing all your income from one day to the next.

Why learn this

It trains your mind to find creative ways out under pressure. The same skill helps you find ways forward for a product.

Key concepts

Contingency plan, creativity under constraints, extreme scenario, habit.

Some entrepreneurs keep a notebook for years with just one question: “if I lost everything tomorrow, how would I feed my family legally?”. The answers become more and more concrete and creative: sleeping near universities, buying a suit at a thrift store to get into events with free food, shaving to look like any other attendee.

Why such an extreme exercise helps

Thinking about the worst-case scenario takes an abstract fear and turns it into a list of actions. After a few years, you discover that there’s almost always a way out, as long as you look creatively at the resources available.

The same mindset, applied to products, leads to very useful questions: “if I had zero budget for ads, how could I get the first users?”

Try it now

Create your product version of the contingency journal with help from an AI assistant (Claude, ChatGPT or Codex).

Quero praticar pessimismo produtivo sobre meu produto.

Produto: <descreva em uma frase>
Público: <quem usa>
Recursos que tenho hoje: <tempo por semana, dinheiro, contatos, habilidades>

Imagine os três piores cenários realistas para os próximos 90 dias
(ex.: ninguém se cadastra, um concorrente grande lança algo igual, eu fico sem tempo).
Para cada cenário, liste 3 saídas criativas que usem SÓ os recursos que eu já tenho.
Nada que exija investimento novo. Seja concreto: ação, prazo, custo.

How to check: You should get 3 scenarios × 3 ways forward. If any option requires money you don’t have, ask the assistant to replace it. Save the result in a note and review it in 30 days.

Start small

It doesn’t have to be a journal you keep for years. Open a note called “contingency” and add one line each week. The value comes from repeating it.

The five reasons it WON’T work

A purely negative exercise

What it is

List the five strongest reasons your idea will fail, without softening any of them.

Why learn this

It’s the cheapest way to find the weak point before the market finds it for you.

Key concepts

Distribution, copying, addressable market, individual pain point, local scope.

Take your idea and answer without holding back: what are the five strongest reasons this won’t work? The list below shows the ones that come up most often. Use it as a starting point, not as an answer.

The most common reasons

1. I don't have distribution. The problem most people face. No one will know the product exists. Without an audience, partners, or a channel, even the best product dies in silence.
2. Someone could copy it by recording the screen. The clone test. Someone records a video of themselves using your product, gives it to a coding assistant, and says, "Make it the same." If that's enough, you don't have a product; you have a demo.
3. The pain might be mine alone. Unknown market. The problem exists in the way you work, but you don't know whether hundreds, thousands, or millions of people feel the same way.
4. I don't know the size of the market. Uncertain TAM. There could be five potential customers in the world, or five million. Without an estimate, you don't know whether it's worth the effort.
5. It might be local only. Geographic scope. The need may exist in your city or region and not make sense elsewhere.

Terms in this topic

• TAM: Total Addressable Market. The total market you can serve. How many people or businesses could, in theory, buy your product.
• Distribution: Everything that gets the product to the people who need it: audience, partners, search, referrals.

Don't skip this exercise

It's uncomfortable to write down why your idea will fail. That's exactly why almost no one does it, and almost everyone discovers these reasons too late, after months of work.

The five reasons it could work

The same rigor, now for the positive side

What it is

List the five concrete reasons why the idea could work, based on resources and channels you actually have.

Why learn this

The positive side shows you where to start. But it needs to be as honest as the negative side.

Key concepts

Your own channels, commission, groups and forums, early feedback, minimum cost.

Now do the reverse exercise. What are the five strongest reasons why this could work? The rule is the same: no wishful thinking, only facts and resources you have today.

• Partners who would promote it. People you know with an audience on social media who would talk about the product in exchange for a commission on each sale.
• A channel of your own. Even a small one: an email list, a profile, a group of former customers.
• Test groups. Forums and groups where your audience already gathers and welcomes useful tools.
• Early feedback. A way to find out in days, not months, whether the product is on the right track.
• Low cost to test. The chance to find out whether it’s worth it while spending almost nothing.

A strong reason in favor

• “I have three friends with 20,000 followers in this niche who are willing to promote it.”
• “I already have 200 former customers with this problem.”
• “I can test this with a group of 5,000 people this week.”

A weak reason in favor

• “Everyone will want this.”
• “If it goes viral, I’ll hit it big.”
• “The product is much better than the others.”

Cross-check the two lists

For each reason against, see whether any reason in favor neutralizes it. “No distribution” is neutralized by “partners who promote it.” Risks without a match on the other list are the ones that need a plan.

The minimum bet

Spend as little as possible to find out if it’s worth it

What it is

Design the cheapest test possible that answers the riskiest question about the product.

Why learn this

With AI, testing has become extremely cheap. Spending months before testing has become an expensive, avoidable mistake.

Key concepts

Small bet, minimum test, risky assumption, cost of learning.

The last part of productive pessimism is to ask: how can I find out if I’m right while spending as little as possible? Not the minimum needed to build the whole product, but the minimum needed to answer the most dangerous doubt.

Examples of minimum bets for different types of doubts

• Does anyone want this? · Minimum bet: Simple page with a waitlist, shared in 2 groups · Sign of success: 30+ sign-ups in one week
• Will anyone pay? · Minimum bet: Preorder button with a discount · Sign of success: 3+ real payments
• Can people understand it on their own? · Minimum bet: Prototype used by 5 people without help · Sign of success: 4 out of 5 reach the result
• Can I reach the audience? · Minimum bet: One post in a niche forum · Sign of success: Replies and direct messages

Be careful with a bet that doesn't answer anything

Opinion surveys ("would you use it?") are cheap, but they prove almost nothing. People say yes to be polite. Prefer tests that require them to take action: sign up, pay, or use it.

AI pre-mortem

Ask the model to kill your idea before the market does

What it is

Use an AI assistant to conduct a full pre-mortem: imagine the product failed in a year and reconstruct why.

Why learn this

The model isn't attached to your idea. With the right instructions, it finds weaknesses you don't want to see.

Key concepts

Pre-mortem, devil's advocate, structured prompt, mitigation plan.

You can bring everything from this module together in one AI exercise. The technique is called a pre-mortem: instead of asking "what could go wrong?", you say "it has already gone wrong" and ask the model to reconstruct the story of the failure. Changing the verb tense opens the door to much more concrete criticism.

Try it now

Run a full pre-mortem of your idea in Claude, ChatGPT, or Codex, weighing reasons against it, reasons in favor, and minimum bets.

Aja como um investidor experiente e cético. Faça um PRÉ-MORTEM do meu produto.

Produto: <o que faz, em uma frase>
Para quem: <público específico>
Como ganha dinheiro: <modelo de cobrança e preço>
Recursos que tenho: <tempo, dinheiro, audiência, contatos>

Parta do fato: estamos daqui a 12 meses e o produto FRACASSOU.

1. Liste as 5 causas mais prováveis do fracasso, da mais para a menos provável.
   Considere obrigatoriamente: distribuição, facilidade de cópia (alguém grava a
   tela e pede a uma IA para refazer), tamanho real do mercado, se a dor é só minha,
   e se um gigante poderia lançar algo parecido.
2. Liste as 5 razões mais fortes para ter dado CERTO, usando só os recursos que informei.
3. Para cada causa de fracasso, proponha uma aposta mínima (menos de 7 dias e
   menos de R$ 200) que testaria se aquele risco é real.
4. Termine com um veredito de uma frase: seguir, ajustar ou abandonar, e por quê.

Não suavize. Se a ideia for fraca, diga.

How to check: The response needs to have all four numbered parts and a clear verdict. If the model praises it too much, reply, "You're being kind; try again as if your own money were at stake." Save the result: it becomes your test plan for Track 2.

Run it twice

Run the same prompt in two different models, or twice in the same model in new conversations. Causes that appear in both responses are the ones you should test first.

Good use of the pre-mortem

• Run it before writing code.
• Turn each cause into a low-cost test.
• Repeat when the idea changes.

Bad use of the pre-mortem

• Keep asking until the model agrees with you.
• Read it, find it interesting, and carry on as before.
• Use it as an excuse to never get started.

Module 1.3 summary · Productive pessimism

• Productive pessimism: list risks so you can take action, not give up
• Contingency journal: practice finding creative ways forward with resources you already have
• Five reasons against: distribution, copying, pain only you feel, uncertain market, local scope
• Five reasons in favor: only current facts and resources, with the same rigor
• Minimum bet: the cheapest test for the most dangerous doubt
• AI pre-mortem: declare the failure and reconstruct the causes with a skeptical model

Is this necessary?

Did you build it because you could, or because the market is asking for it?

What it is

The first opportunity question: Does the product come from real demand, or just the excitement of being able to build it?

Why learn this

Today, you can build almost anything in an afternoon. That’s why “I can make it” is no longer a good argument. Most ideas people talk about don’t solve any real pain.

Key concepts

Real demand, technical excitement, pain, a market that “is shouting,” builder’s idea.

There’s a kind of idea that comes from the builder’s side: “I found out I can make X with AI, so I’ll make an X app.” The missing question is from the perspective of the person who will use it: Is anyone suffering because they don’t have X?

Anyone who talks with many entrepreneurs notices an uncomfortable pattern: the vast majority of ideas they present, something like nine out of ten, don’t address a pain point. They’re nice things, well made, sometimes beautiful. But nobody wakes up in the morning needing them.

Signs the market is asking for it

• People already pay for workarounds (spreadsheets, freelancers, consulting) to solve this.
• You hear the same complaint from people who don’t know one another.
• The problem measurably costs money, time, or sleep.
• Someone asks “when will it be ready?” before you offer it.

Signs it’s just your own excitement

• The explanation starts with the technology, not the problem.
• You have to convince people they have the problem.
• The compliments are “that’s cool,” not “how much does it cost?”
• No user came back after the first test.

The “How much does it cost?” test

Show the idea to five people in the right audience. If no one asks the price, timeline, or “can I try it now?”, you heard politeness, not demand.

Solve based on the pain

Without pain, no one has a reason to change their habits

What it is

Build around someone’s specific pain instead of building just for the sake of solving something.

Why learn this

Adopting a new product takes work: signing up, entering card details, learning how to use it. Only someone with enough pain will pay that switching cost.

Key concepts

Pain, switching cost, urgency, intensity, frequency.

If you don’t solve a problem based on the pain, you’re solving just for the sake of solving. Then no one is in enough pain to justify using your product. This may sound obvious, but it’s the thing people forget most when building becomes easy.

Why pain fuels adoption

Every switch to a new tool has a cost: learning the interface, moving data, convincing your business partner, entering card details. Users pay that cost only when the pain of staying as they are is greater than the pain of changing.

That’s why “cool” products get lots of sign-ups but little use. Signing up costs one click; daily use requires pain that keeps coming back.

Four dimensions to assess whether a pain point can sustain a product.

• Frequency · Question to measure it: How often does this happen? · Strong answer: Every week or every day
• Intensity · Question to measure: How much does it cost when this happens? · Strong answer: Money, a lost customer, a sleepless night
• Urgency · Question to measure: Does this need to be solved now? · Strong answer: Yes, there’s a deadline, a fine, or a risk
• Awareness · Question to measure: Does the person know they have this problem? · Strong answer: Yes, and they’ve already looked for a solution

Trap: the pain only you feel

A problem may be real for the way you work and rare for everyone else. Before building anything, confirm that people beyond you and your close friends have this pain too.

Will it still be worth having in 3, 6, and 12 months?

Live three months in the future

What it is

Project the idea into the near future and ask whether it will still have value as models and tools advance.

Why learn this

What requires a product today may become a free feature or a single request to a model in a few months. Building it means building something with a short shelf life.

Key concepts

Shelf life, “vibe-codable,” one prompt, feature vs. product, 3/6/12-month horizon.

A good discipline is to try to live three months in the future. When a new product catches your eye, the question isn’t “is this cool?” but: how easy will it be to add this same thing anywhere else three months from now?

Terms in this topic

• Vibe coding: Building software by describing what you want to AI, which writes the code. “Vibe-codable” means something you can make this way with little effort.
• One prompt: When an entire feature comes from a single request to the model. If your product fits in one prompt, it isn’t a product; it’s an instruction.
• Clone with a tiny difference: A product that repeats an existing one with one extra detail. The original usually absorbs that detail within weeks.
• Shelf life: The time during which the product still solves something the free alternative doesn’t.

The three horizons exercise

1. Three months from now. Can you already do this with a coding agent in an afternoon? If so, your window is short.
2. Six months from now. Will one of the major assistants offer this as a built-in feature? If so, your audience will get for free what you charge for.
3. Twelve months from now. Will the problem still exist, even with much better models? If it changes along with the technology (like security), that’s a good sign.

Why ignore most launches

This filter explains why it’s worth ignoring most of the new products that appear every week. If the difference is a feature anyone can add in a month, it isn’t worth your attention, either as a user or a competitor.

Case study: security for apps built with AI

Boring, nobody likes it, and it's an endless game of cat and mouse

What it is

An example of a niche deliberately chosen because it's unappealing and renews itself: protecting applications built with AI.

Why learn this

Shows the earlier criteria in practice: a real pain point, a problem that isn't solved once and for all, and one that grows along with the technology.

Key concepts

Cybersecurity, cat and mouse, growing attacks, open models without safeguards, defense with frontier models.

Imagine someone choosing a niche for their next SaaS product. Instead of something flashy, they choose security for applications built with AI. There are three reasons, and none is "because it's beautiful."

• Nobody likes it. Most people avoid the topic of security. That means less competition from people who only want to do things that get likes.
• It's not sexy. It won't get you a viral video saying "look what I made in 10 minutes." People who get into it do so out of necessity, and they stay.
• An endless game of cat and mouse. Every advance in defense leads to an advance in attacks. The problem is never "solved for good."

Why it's a game of cat and mouse

Frontier models are getting better at defending. At the same time, open models running locally without safety safeguards, often trained by copying the behavior of closed models, are getting better at breaking in.

And there's one detail: frontier models have safeguards that limit their help to ordinary people with offensive security tasks. That leaves room for specialized products that strengthen the infrastructure of people who aren't large companies.

What's good about this niche

• Demand grows with AI's own progress.
• Customers feel afraid, and fear is a strong pain point.
• Solving it requires ongoing updates, which supports a subscription.
• Data about vulnerabilities found becomes an asset and content.

What it requires of you

• Credibility: nobody trusts security to someone who seems like an amateur.
• Constant updates, without ever "finishing" the product.
• Explaining a boring topic simply.
• Real responsibility when something fails.

The information gap is a pain point

The fear of what you don't know you don't know

What it is

Many pain points aren't about lacking a tool, but lacking knowledge: the person knows there's a risk, but doesn't know how to assess it.

Why learn this

This kind of pain point is strong because it comes with insecurity. And it's a problem people don't want to spend time or tokens on, even when they know they could.

Key concepts

Information gap, insecurity, hiring a specialist, time and tokens, outsourcing the worry.

Think of someone in the middle of a contract with a client who suddenly realizes the delivered system may have security flaws. Their first reaction is to think they need to hire an expensive specialist. The discomfort is not technical; it’s emotional: they don’t know what they don’t know.

Even people who could solve it with help from an AI agent often prefer not to spend time or tokens on it. It’s the kind of problem they want to hand off to someone they trust and forget about. That “I want to outsource the worry” is a great product signal.

How to find information gaps

Look for phrases like “I don’t know if I’m doing this right,” “I’m afraid I may be exposed,” and “Do I need a specialist?” They show up in forums, groups, and comments, and point to pain that no one has organized yet.

Unappealing problems beat attractive ones

Flashy things attract competition; boring things attract customers

What it is

Choose unglamorous problems that few people want to solve over flashy problems everyone is tackling at the same time.

Why learn this

“Sexy” problems make for great demos, so they get crowded. And some popular niches are already handled well by AI chat itself, without needing SaaS.

Key concepts

Sexy problem, boring problem, saturation, content creation, personal process.

People love chasing attractive problems because they look good: you can record an impressive video in ten seconds. The result is predictable: dozens of nearly identical products competing for the same attention.

A classic example of a bad niche today is content creation automation. There are a thousand ways to do it, and most people already use ChatGPT, Claude, or both for this. Plus, content creation is a personal process: a generic tool never works quite the way someone likes.

Try it now

Ask an agent (Claude Code, Codex, or chat) to honestly evaluate your idea using the questions in this module.

Aja como um investidor cético avaliando uma ideia de produto.

Ideia: <descreva seu produto em 2 frases>
Público: <quem usa e em que situação>
Como a pessoa resolve isso hoje: <ferramentas, planilhas, freelancer...>

Responda, com nota de 0 a 10 e uma frase de justificativa cada:
1. Isso é necessário ou foi construído porque era possível?
2. Qual é a dor, em frequência, intensidade, urgência e consciência?
3. Isso ainda vai merecer existir em 3, 6 e 12 meses? Poderia virar um único prompt?
4. É um problema "vistoso" (lotado) ou "chato" (com pouca concorrência)?
5. Há uma lacuna de informação (medo do que o cliente não sabe)?

Termine com: "Vale continuar?" (sim / não / sim, se...) e as 3 perguntas que eu deveria fazer a clientes reais esta semana.

How to check: The answer should include 5 scores with reasons and a clear decision. If all the scores are high, ask: “now be tough and find the 3 biggest flaws.”

Be wary of a high score

Language models tend to be agreeable. Run the same prompt asking explicitly for the role of a critic, and compare the two answers.

Module 2.1 summary · Need and shelf life

• Need: Building something because it’s possible is not a reason; build it because someone needs it.
• Pain: Only pain is worth the cost of changing a habit.
• Shelf life: Think 3, 6, and 12 months ahead: if it becomes a prompt, forget it.
• Cat and mouse: Problems that renew themselves with technology can sustain products.
• Information gap: Fear of what you don’t know is a strong, often overlooked pain point.
• Boring beats flashy: Less competition and customers who stick around.

Could a giant clone you overnight?

Credibility in the niche and the charm of the independent product

What it is

Assess whether a large company with plenty of resources could wake up tomorrow and launch your product, and whether it would have credibility in that market to take your customers.

Why learn this

Being copied by a giant is the biggest risk for a small product. But not every giant has the right reputation in every niche.

Key concepts

Cloning, credibility in the niche, independent product, frictionless experience, landing page designed for the person.

There are companies that, if they wanted to, could clone your product in a week. The full question has two parts: could they? and would they have credibility in that market to win your customers?

Sometimes the answer to the second question is no. Your advantage may be that you are the independent, modern product people want to bet on. Easy to use, without red tape, with a page that feels made for the person, not for a corporation's purchasing department.

You are better able to withstand a clone when

• The niche is too small to interest the giant.
• The giant has no reputation with that audience.
• Your product carries an identity people want to support.
• You solve the specific case, not the generic one.

You are exposed when

• Your product is a natural built-in feature of the giant's product.
• The audience already uses the giant every day.
• Your only advantage is having arrived first.
• Your page looks like any corporation's page.

Pocket question

"If the biggest company in my industry launched this tomorrow, why would a customer stay with me?" If you don't have a one-sentence answer, you don't have protection yet.

The built-in feature test

What if this became an app inside the assistant the customer already pays for?

What it is

Imagine your product as a built-in feature, an app, or a feature inside ChatGPT, Claude, or Gemini, accessible with one click and no new subscription.

Why learn this

If an 80% good version appears inside something the customer already pays for, they don't need to find you, trust you, or enter their card. That's the concrete risk.

Key concepts

Built-in feature store, 80% of the way there, zero adoption cost, "is it worth a prompt for them?", toss them a bone.

Do the exercise: tomorrow, one of the big AI labs adds your product as an app to its assistant's built-in feature store. The user clicks once and uses it. They don't need to sign up for anything else, know your brand, or risk their card. Even if it solves only 80% of the problem, many people will think that's enough.

The question that balances the fear

There's another side: is it worth a giant lab's time and resources, even if all it takes is sending a request to its strongest model, to solve your micro-problem?

Often, no. They'd rather leave that field open, launch something generic, and "toss users a bone" without solving the whole problem. The space between the bone and the complete solution is where small products live.

How to read the risk of being swallowed by a built-in feature.

• Generic feature every user wants · Built-in feature risk: High · Takeaway: It will become a native feature; avoid it
• A niche case with specific rules · Built-in feature risk: Low · Takeaway: The giant throws you a bone and moves on
• Requires customer data or integrations · Built-in feature risk: Medium · Takeaway: It depends on how deeply it fits
• Requires taking responsibility (safety, legal) · Built-in feature risk: Low · Takeaway: Big companies avoid taking on this risk

Don’t confuse “they haven’t done it yet” with “they won’t do it”

A feature being absent today is not protection. Protection is the structural reason it isn’t worth it for them to do it well.

The intrinsic value of data

Sometimes software is just the machine that produces the asset

What it is

Ask whether, even if the product is surpassed, the data it produced has value of its own—enough for someone to buy it.

Why learn this

Companies have been sold for fortunes not because of their software or users, but because of the data only they had.

Key concepts

Intrinsic asset, dataset, acqui-hire, hard-to-generate data, data buyers.

A real example helps. A young, low-key founder in a T-shirt and sandals had just sold the company for nine figures. The company was in the music industry. It wasn’t bought because of its software or, especially, its number of users. It was bought for its data: a hard-to-generate collection that you couldn’t buy from any data broker.

Terms in this topic

• Dataset: An organized collection of data, such as records, text, audio, or interactions, used for analysis or to train models.
• Acqui-hire: When a company buys another company mainly to hire its team. In the example, part of the team joined to integrate the product.
• Intrinsic asset: Something that has value in itself, whether or not the product succeeds.
• Data broker: A company that buys and resells data. If your data can’t be bought from one, it’s rare.

The data market is active

• Some companies advertise that they pay up to US$ 2 milhões for sufficiently large sets of business data.
• They select and organize this data to resell it to AI labs.
• As a result, even if your SaaS doesn’t take off, its byproduct may be sellable.

Be careful with privacy

Valuable data is an asset only if it was collected with consent and within the law (in Brazil, the LGPD). Data collected improperly is a liability, not an asset.

Design the product whose byproduct is an asset

Think about data from day one

What it is

Plan from the start what data product use will generate, how to store it, and how it can become valuable.

Why learn this

Good data doesn’t appear by accident. If you don’t design for it from the start, you’ll lose the history that could give the business value even in the worst-case scenario.

Key concepts

Data as a byproduct, anonymization, structure, consent, plan B.

If data can be the asset, it needs to be designed as one. That doesn’t mean collecting everything. It means collecting what is rare, structured, and authorized.

Four steps to turn use into an asset

1. Identify the rare data. What does your product see that no one else sees? For example, a security product sees which vulnerabilities appear across thousands of different apps.
2. Structure it from the start. Store it in a consistent format (categories, dates, context). Messy data is worth little.
3. Make sure you have consent. Clear terms of use, anonymization, and compliance with the LGPD. Without these, the data can’t be sold or used.
4. Use the data now. The same data can power reports, content, and product improvements before it has value as an asset for sale.

Data that looks like an asset

• Rare: only your product generates it.
• Structured and labeled.
• Collected with consent.
• Grows with use, at no extra cost.

Data that looks like trash

• Generic: anyone can collect it.
• Scattered in logs with no standard format.
• Collected without permission.
• Requires manual work to exist.

Too easy to build is a problem

If it took half an hour, someone can make the same thing in half an hour

What it is

Treat ease of building as a warning sign: if the product is trivial to make, it’s trivial to copy.

Why learn this

In a world where anyone can build anything, difficulty has become a form of protection.

Key concepts

Difficulty as a moat, one-shot, iteration, rare insight, accumulated effort.

If it’s too easy to build, there’s usually a problem. Unless you’ve had a brilliant idea no one in human history has ever had—which is very unlikely—something narrowly focused that took half an hour to make will rarely work.

There needs to be some pain point or iteration cycle that, even using the strongest model available at the highest effort level, can’t be done in a single request.

The "one request" test

Describe your product in one paragraph and ask a coding agent to build it. If the result is 80% similar within an hour, your differentiator isn’t in the software. It needs to be somewhere else: data, niche, distribution, or nuance.

The nuance that makes it feel instant is the moat

What lies behind the simple

What it is

The real value is in the invisible work that makes a result feel easy and immediate for the user.

Why learn this

The user sees a text box and a button. A competitor copies the box and the button. What they can't copy is the chain of decision steps between what goes in and what comes out.

Key concepts

A proprietary chain of steps, nuance, end-to-end design, an instant experience, care.

Some products seem almost too simple: you paste something, click, and the result appears. It seems trivial. But everything that happens behind the scenes to make it feel instant is the real value: the nuance, care, and proprietary design of the path from what goes in to what comes out.

Where nuance usually lives

• Handling unusual cases that only become clear through use.
• The order and combination of processing steps.
• Checks that prevent incorrect answers.
• Speed achieved through engineering, not luck.

What nuance is not

• Changing the interface color.
• Changing the main prompt text.
• Adding a feature anyone can add.
• Using the most expensive model without thinking.

Try it now

Ask Claude Code to analyze your project and point out where defensible nuance exists—or is missing.

Analise este pasta do projeto como se você fosse um concorrente tentando copiar o produto.

Contexto: <o que o produto faz, em 2 frases>

1. Liste o que você conseguiria replicar em menos de um dia.
2. Liste o que exigiria semanas ou meses (regras, casos especiais, cadeia de etapas, dados).
3. Aponte 3 pontos onde o produto parece simples para o usuário mas depende de lógica não óbvia.
4. Sugira 3 melhorias que aumentariam essa nuance difícil de copiar, sem aumentar a complexidade para o usuário.

Responda em tabelas curtas. Não altere nenhum arquivo.

How to check: If list 1 covers almost everything and list 2 is empty, your product still has no technical moat. Use answer 4 to plan next week.

Nuance is not complexity for the user

The goal is to hide the work, not expose it. If nuance forces the user to configure twenty options, it has become friction, and friction pushes customers toward the simpler alternative.

Module summary 2.2 · Giants, data, and difficulty

• Giants: Ask whether they could replace you and whether customers would trust them to do it.
• Built-in feature test: A built-in feature that's 80% good is the most concrete risk.
• Throwing a bone: Big companies often leave the micro problem without a complete solution.
• Data: The byproduct may be worth more than the product; design for it from the start.
• Difficulty: Easy to build means easy to copy.
• Nuance: The invisible work that makes it feel instant is the moat.

Skate to where the puck is going

Anticipate instead of chase

What it is

The hockey analogy: don't skate to where the puck is; skate to where it's going. For a product, that means building for the market as it will be in a few months.

Why learn this

Those who build for the current state arrive after the opportunity has passed. The pace of AI change makes that window much shorter.

Key concepts

Anticipation, cannibalization, window of opportunity, extrapolating trends.

In hockey, a good player doesn't chase the puck. They watch who's carrying it and anticipate where it will be, so they can get there first and, hopefully, score. Products work the same way.

The practical question: if the industry is moving toward a point where your idea could be swallowed up, will your customer notice? And when?

Those who skate to where the puck is going

• They read the signs: new pain points, falling costs, changing habits.
• They ask what will become easy in 3-6 months and what will stay difficult.
• They build the part that won't become a commodity.
• They arrive when the customer starts looking.

Those who chase the puck

• They copy what's already successful today.
• They build on a limitation that the next model will solve.
• They arrive when the market is already crowded.
• They compete on price alone.

A simple habit

Every week, write down three things that became easier with AI. In a month, you'll have a list of what is about to become a commodity and what hasn't yet.

Arbitrage of naivety

The customer still doesn’t know they could do it themselves

What it is

The window of time when your audience hasn’t yet discovered they could build on their own, with AI, what you sell.

Why learn this

Even if the solution can be replicated, many customers will only realize that in 6 to 12 months. That time is an opportunity, as long as you have a plan for what comes next.

Key concepts

Arbitrage, market naivety, 6-12 months, scale, sell, or pivot.

Terms in this topic

• Arbitrage: Earning from a temporary difference between two places or moments. Here, the difference is between what’s already possible and what the customer knows is possible.
• Market naivety: The audience’s lack of awareness of what AI can already do. It’s not the customer’s fault: no one can keep up with everything.
• Pivot: Change the product’s direction using what has already been built (customers, data, brand).

Many audiences won’t discover they can “vibe-code” a particular solution for another six or twelve months. Until then, there’s an arbitrage of naivety: you sell something they could technically make themselves, but they don’t know that.

The plan for when the window closes

1. Scale. Grow enough for your brand, data, and distribution to protect you when naivety fades.
2. Sell. Negotiate the business (or the data) while it’s still worth a lot.
3. Pivot. Use customers and what you’ve learned to tackle the next problem that hasn’t become a commodity yet.

Arbitrage isn’t deception

Customers pay for convenience, trust, and time saved, and that’s legitimate. Hiding information or exaggerating how difficult something is to keep them dependent isn’t.

Slow-moving industries mean longer windows

People who work with the physical world don’t have time for digital

What it is

Selling to traditional industries that make a lot of money doing one thing well and don’t keep up with AI developments.

Why learn this

In these industries, the arbitrage of naivety lasts much longer. Even with much stronger models, they’ll still be discovering the basics.

Key concepts

Slow-moving industry, oil and gas, metals, physical world, slow adoption cycle.

Think of industries such as oil and gas, precious metals, mining, traditional agriculture. Even if a model ten times better came out tomorrow, many of them would only be discovering that you can connect an AI assistant to email.

These companies make a lot of money doing one thing well in the physical world. They’re focused on metalworking processes, logistics, and workplace safety, not on which model launched this week. If even people who work in technology can’t keep up with everything, imagine people who work in another field.

Why it’s worth targeting slow-moving industries

• They have plenty of money and costly processes.
• They won’t build it themselves.
• They value relationships and support.
• There’s less competition from startups.

What they’ll require

• Longer sales cycles.
• Industry language, not technology language.
• Security, compliance, and contracts.
• Patience for implementation.

Translate, don’t impress

In these industries, no one wants to hear “agent with a frontier model.” They want to hear “the report that used to take two days now takes twenty minutes, in the same format as always.”

Cat and mouse versus saturation

Problems that renew themselves versus problems that run their course

What it is

Distinguish between problems that are never fully resolved (and support a subscription) and problems that can be solved well enough and become saturated.

Why learn this

Customer retention and lifetime value depend on this. A saturated problem becomes free; a problem that keeps renewing itself can sustain a subscription.

Key concepts

Recurring revenue, customer lifetime value, saturation, open source, “Why would I pay?”

Think about user interface component libraries for developers. Most of the content is free: you copy a prompt or a snippet of code, and you’re done. There are hundreds of alternatives, many of them open source. Why would anyone pay? It’s hard to make money from a problem that’s been solved well enough and has many alternatives.

Four questions to test whether the problem can sustain a subscription.

• Does the problem keep coming back? · Answer that supports recurring revenue: Yes, every week or whenever the environment changes
• Does the solution need ongoing updates? · Answer that supports recurring revenue: Yes, the world changes and the solution keeps up
• Are there equivalent free alternatives? · Answer that supports recurring revenue: No, or they’re much worse
• Does the customer lose anything if they cancel? · Answer that supports recurring revenue: Yes: history, protection, data, routine

That’s why the most useful question is: what annoyance is worth solving that most people won’t solve on their own? Not the sexy one, but the tedious one that keeps coming back.

Live in the daily routine and make passive progress

Be there without the user having to remember you

What it is

Make the product part of the customer’s routine, working in the background and delivering progress they couldn’t make on their own.

Why learn this

The factor that keeps a customer around most is being part of their daily routine. Products that depend on users remembering to open them get forgotten.

Key concepts

Daily routine, background, proactive, symbiosis with automation, passive progress, data the user already has.

It’s very hard to make a product that runs in the background and that the user wants to have there. Two examples from the world of development show the way.

• Error monitor (e.g., Sentry). It watches the application and alerts you when an error appears for users. The developer doesn’t need to remember to open anything: the alert reaches them.
• Code reviewer (e.g., CodeRabbit). It analyzed a huge number of code changes and learned patterns of errors. After the agent writes the code, it warns you: “This will break in production. Send it back for fixes.”

Notice the second example: it exists in symbiosis with the tools that are automating development itself. Many people don’t even look at the code the agent writes anymore; they “close their eyes and pray.” The product makes that prayer safer. It doesn’t compete with automation; it builds on it.

Example: a bank statement becomes an investment thesis

Imagine an app where you connect your bank account through a provider with security certifications. Every day, it reviews your spending and suggests investments that fit what you buy, with an explanation.

If you pay for almost everything with a credit card, and notice that everyone does the same, it makes sense to consider payment companies. The app doesn’t invest for you: it gives you a list and explains why, using data you already had but weren’t making use of. What we buy reflects what we value, and that often looks like what similar people value.

Financial recommendations require care

This example is for illustration. A real product in this area needs to comply with local regulations (in Brazil, CVM and the Central Bank) and make it clear that it is not individual financial advice.

A good feeling and becoming a verb

Peace of mind as a product, and the word that enters everyday vocabulary

What it is

Two extra signs of recurring use: the product creates a pleasant feeling (relief, security) and has the potential to become an everyday verb.

Why learn this

A good feeling makes users come back on their own. Becoming a verb is rare, but when it happens, the language itself does the marketing.

Key concepts

A feeling of security, relief, antivirus software, manufactured demand (ethical warning), a brand that becomes a verb.

Remember when running antivirus software was a ritual? The little bar moved along, “no threats found” appeared, and you felt relieved. That warm feeling of security is part of the product. A dull topic like security can create a very good feeling.

The dark side: manufactured demand

There are old reports of providers helping create the problem so they could sell the solution. That isn't a strategy; it's fraud. The legitimate lesson is different: identify a real fear and provide real peace of mind.

The last sign is the rarest: can the product become a verb? “Google it,” “get an Uber,” “send it on Slack.” Nine out of ten products never get there, and that's fine. Still, it's worth asking whether the name and main action are simple enough to become part of everyday speech.

Try it now

Assess how often your idea will be used with the signs from this module and get concrete suggestions for increasing it.

Avalie a recorrência do meu produto.

Produto: <o que faz, em 2 frases>
Público: <quem usa, setor, nível técnico>
Frequência atual de uso esperada: <diária / semanal / mensal / esporádica>

Para cada sinal abaixo, diga "sim", "parcial" ou "não" e justifique em 1 frase:
1. Problema de gato e rato (se renova sozinho)
2. Arbitragem da ingenuidade (quantos meses até o cliente saber fazer sozinho?)
3. Setor atrasado (janela longa)
4. Vive no fluxo diário / em segundo plano
5. Gera progresso passivo com dados que o usuário já tem
6. Entrega uma sensação boa (alívio, segurança)
7. Potencial de virar verbo

Depois proponha 3 mudanças concretas que aumentariam a frequência de uso sem aumentar o atrito.

How to check: You should receive 7 evaluations and 3 changes. If the changes are generic (“improve the marketing”), ask: “give me product changes, with the exact screen or trigger.”

Combine the signs

The strongest products combine several signs: a security monitor for apps made with AI is a cat-and-mouse game, runs in the background, provides relief, and keeps accumulating data. Each sign reinforces the others.

Module 2.3 summary · Use, recurring use, and habit

• Anticipate: Build for where the market is headed, not where it is now.
• Naivety arbitrage: Take advantage of the window, with a plan to scale, sell, or pivot.
• Slow-moving industries: Years-long windows for those who speak their language.
• Cat and mouse: Problems that keep coming back support subscriptions; saturated ones become free.
• Daily routine: In the background, working in sync with automation, with progress happening passively.
• Feeling and verb: Real relief brings people back; becoming a verb is rare, but worth pursuing.

Distribution affects everyone

The finished product nobody saw

What it is

Distribution is the set of ways the right people discover, try, and return to your product.

Why learn this

When anyone can build something in days, the advantage is no longer the code. It's who gets in front of the right people first.

Key concepts

Heartbeat, audience, channel, surface area for luck, early signs.

A pattern repeats in almost every new project: someone spends weeks on the product, publishes it, and hears nothing. No sign-ups, no comments, no complaints. Not because the product is bad, but because nobody knows it exists. The first job after building is to create a heartbeat: any sign that someone outside your circle cares.

Why this got worse instead of better

When building required a team and months of work, few competitors arrived at the same time. Today, with coding agents, dozens of similar products appear in the same week. The result is an inflation of launches: people see "new" all day and start ignoring almost everything.

In this situation, distribution isn't the final step in the plan. You need to think about it alongside the idea. The question "how will the first hundred people find out about this?" needs an answer before the first line of code.

Signs of life

• Strangers sign up without you asking.
• Someone comes back the next day.
• Questions and complaints start coming in (that’s a good sign).
• One person recommends it to another.

Signs there’s no life yet

• Only friends and family tried it.
• All the visits come from your own post.
• No one complains about anything.
• The numbers go up only on launch day.

Pocket question

Before starting any product, write this in one sentence: "The first 100 people will find me through ____". If you leave it blank, you don’t have a plan yet, just an idea.

Reddit and communities

Help first, mention it in passing

What it is

Take part in forums and groups where your audience already talks, building a reputation before you mention the product.

Why learn this

It’s a free channel that doesn’t require followers, and many service businesses have landed contracts from a good answer in a forum.

Key concepts

Karma, self-promotion, social proof, long-term, natural mention.

Communities like Reddit have a measure called karma: points you collect when other people upvote your answers. Accounts with high karma appear more prominently and are less likely to be filtered as spam. In other words, reputation is literally what opens the channel.

Terms in this topic

• Karma: Points you collect from upvotes on your contributions. It works like trust credit.
• Self-promotion: Posting only to promote your own product. Almost every community penalizes or removes it.
• Natural mention: Mentioning the tool you use in passing, in an answer that already solves the person’s problem.
• Subcommunity: A topic-focused forum (on Reddit, a "subreddit") where a specific audience gathers.

The four-step path

1. Map where your audience asks questions. List 5 to 10 forums, groups, or subcommunities where people talk about the problems your product solves. Read before you write.
2. Answer without selling. For weeks, solve real problems. Write complete answers with step-by-step instructions. That’s what builds karma and makes people remember you.
3. Mention it in passing. When the question is exactly what the product solves, end your answer with something like "I use X for that." One sentence, not an ad.
4. Measure and repeat. Use a link with a source parameter to see which answers brought people in. Double down on the forums that work.

Works

• Answer the whole question, even without mentioning the product.
• Show a concrete result ("I did it this way, and this happened").
• Come back to answer questions in the comments.

Burns the channel

• Paste the same link into ten threads.
• Create a new account just to promote your product.
• Reply with "check out my app."

Time frame

This is a medium- to long-term channel. It takes weeks to build a reputation and months to become a steady source of new people. Don’t expect results in the first week, and don’t give up in the second.

Guerrilla tactics and small bets

Think outside the box on a small budget

What it is

Creative, low-cost actions that tap into attention already in the market to bring people to your product.

Why learn this

With a small budget, you can’t compete on ads. You compete with creativity and speed.

Key concepts

Guerrilla marketing, asymmetric bet, Trojan horse, attention window, repetition.

Guerrilla marketing is the name for actions outside the traditional playbook: instead of paying for attention, you position yourself where attention already is. The principle is the asymmetric bet: it costs little if it fails and can pay off big if it succeeds.

Example: the 22-dollar domain

An entrepreneur predicted the name of a product that a large company would launch the next day. They bought a domain with that name for about US$ 22, built a small catalog of assistants for that platform in three hours with a coding agent, and spent another twelve hours polishing it.

If the prediction was right, the site would get some of the traffic from curious people looking for the new product. And inside it, there was a discreet highlight for the entrepreneur’s main SaaS. It’s a traffic “Trojan horse”: ride someone else’s wave to bring people to you.

Most common guerrilla tactics, from least expensive to most work

• Ride an expected launch · Cost: Low · When to use: You know the market well and can anticipate what will happen.
• Reply to a trending conversation · Cost: Zero · When to use: A viral discussion touches exactly on the problem you solve.
• Free tool as a hook · Cost: Low · When to use: A small utility related to the product that people share.
• Cold calls and cold email · Cost: Time · When to use: High-ticket sales; it works, but takes a lot of repetition.

Good bets

• A cost you can accept losing without pain.
• Connected to your real audience.
• Quick to build and measure.

Bad bets

• Misleading users about what they’re accessing.
• Using someone else’s brand in a way that creates legal confusion.
• Spending weeks on an action that depends only on luck.

Think outside the box, but measure

Cold calls and cold email work; entire agencies make a living from them. But in a world where everything is about being first in front of people’s eyes, short windows of attention tend to bring in more per dollar invested. Make several small bets and track which one brought people in.

Two-sided affiliate

Commission for the person who refers, credits for the person who uses it

What it is

A referral program where the person who shares earns recurring money and also gets usage credits for the product.

Why learn this

Turns your users into a distribution channel, with an incentive for them to keep referring others.

Key concepts

Affiliate, recurring commission, credits, monthly recurring revenue, double incentive.

A standard affiliate program pays a commission to someone who brings in a customer. The two-sided model goes further: the person who refers earns a commission and also gets credits, a discount, or free months of the product. The incentive is no longer just money. It also means "use it for free."

Terms in this topic

• Affiliate: A person who promotes a product with their own link and gets paid for each customer they bring in.
• Recurring commission: A percentage paid every month as long as the referred customer keeps paying, not just on the first sale.
• MRR: Short for monthly recurring revenue: how much the product earns each month from subscriptions.
• Usage credit: A balance within the product (minutes, generations, queries) that the user spends instead of paying.

An example of how this works

A user of a voice dictation tool recommended the product to their audience. After enough referrals, they earned enough credits to keep their subscription free until 2027. They don’t use the tool as much anymore, but they keep recommending it because each new referral keeps their account free.

For the company, the cost of these credits is small compared with the value of the customers who sign up. For the person making referrals, there’s a concrete reason to keep talking about the product.

Do the math before setting the percentage

• If the monthly subscription is R$ 50 and you pay 40% recurring commission, each referral costs you R$ 20 per month for as long as they stay.
• If it costs R$ 10 to serve that customer, you’re left with R$ 20 in margin. You’re still making a profit.
• Usage credits cost what it takes to provide the service, not the list price. That makes them an inexpensive incentive.

A program that works

• A link that’s easy to find in the product.
• A panel showing how much the person has earned so far.
• A bonus for people who sign up through the link, too.

A program nobody uses

• Separate registration with lots of paperwork.
• Payment only after reaching a high amount.
• Commission only on the first sale.

Practical tip

Show the invitation right after the user gets a good result (“your report is ready”). That’s when they’re most likely to show it to someone.

“Blood in the streets” moments

Show up when people are ready to pay attention

What it is

Use spikes in public attention around the problem you solve to offer your solution generously.

Why learn this

Many small products grow slowly, then suddenly take off when everyone starts caring about the problem.

Key concepts

Timing, a spike in attention, an expanded free plan, goodwill, a hockey-stick growth curve.

There’s a classic saying in finance: “buy when there’s blood in the streets”. It means the best opportunities appear during a panic. For a product, the idea is: show up in a big way when the problem you solve makes the news.

Example: a wave of attacks on apps

Imagine a security product for apps built with AI. One day, several apps are hacked, a well-known AI service goes down, and the story makes the headlines. For a few weeks, everyone is worried about security.

At that moment, the product announces: “we’ve expanded the free plan by 10 times; anyone can check their app now, at no cost.” The result: goodwill, new users, more data (if data is valuable to the product), and a positive impression of the brand.

What happens during these spikes

• Small SaaS businesses that were earning about 600 per month reached 7, 15, or 20 thousand in monthly revenue within a few months.
• The jump rarely comes from a paid campaign. It comes from being ready when attention arrives.
• A free way in is what turns curiosity into use.

How to prepare before the spike

1. List the events that make your problem urgent. Data leaks, a change in the law, a major service going down, a competitor’s launch. Write them down before they happen.
2. Get the offer ready. Have the text, a page, and an expanded plan setting ready to turn on in minutes.
3. Be loud that day. Post, reply to conversations, send it to your list. The peak lasts days, not months.

Opportunism has its limits

Helping someone in a moment of fear is different from exploiting that fear. Offer something that truly helps, with generosity. An alarmist ad on top of a tragedy destroys the trust you wanted to earn.

Accelerators and content creators

Products other people want to show

What it is

Two shortcuts to distribution: get into an accelerator on the strength of your idea, and make a product content creators want to promote.

Why learn this

When organic growth doesn’t take off, an accelerator offers connections and money; content creators are always looking for simple new things to show.

Key concepts

Accelerator, early traction, short video, 15-second demo, keyword in the comments.

If organic traction isn’t coming, ask yourself this: Is the product good enough to get into an accelerator? Programs like Y Combinator, or accelerators in your area, accept ideas with merit and some early traction, even if it’s just a handful of users.

Content creators are a channel

On short-video platforms, the posts that work best show a tool, a project folder, or a feature few people know about. Creators are always looking for these new things to share with their audience.

The products they choose have one thing in common: they can be demonstrated in seconds. “Paste your link here, click here, see what happens. I did this for free in 15 seconds. Comment BANANA and I’ll send you access.”

Easy-to-show product

• A single text box.
• A visual result in seconds.
• Anyone can understand it without an explanation.

Hard-to-show product

• Requires registration, configuration, and integration before the first result.
• The value only shows up after weeks.
• Needs a paragraph to explain what it does.

• Accelerator. Bring your idea, whatever traction you have, and a clear description of the problem. The merit of the idea counts, even with few users.
• Content creator. Send free access and a ready-to-use 15-second worksheet. Make their work easier.
• Your own videos. Short-video platforms don’t require followers. Even a new account can have a video with tens of thousands of views.

Quick test

Record a 20-second video of yourself using the product on your phone. Show it to someone who doesn’t know the project. If they understand what it does, the product is ready for this channel.

Module 3.1 summary · Distribution

• Distribution is part of the idea: Building has become cheap; getting noticed has become expensive.
• Communities: Help first, build a reputation, mention it in passing.
• Guerrilla tactics: Small, asymmetric bets that make use of existing attention.
• Two-sided affiliate: Recurring commissions + credits turn users into a channel.
• Blood in the streets: Generous offers ready for spikes in attention.
• Easy to show in 15 seconds: What fits in a short video can become a channel for creators and accelerators.

Positioning in one line

“Antivirus for apps made with AI” × “the Uber of X”

What it is

The short sentence that says what the product is, who it’s for, and why it matters, using something the audience already understands.

Why learn this

A clear sentence takes little effort to understand. The less effort it takes, the more people understand it.

Key concepts

Positioning, analogy, familiar category, one-line phrase, cliché.

Compare two phrases. The first: "we’re an AI-powered security platform for modern developers". The second: "antivirus for apps made with AI". You understand the second in a second because it uses a category everyone knows (antivirus) for a new problem (apps generated by AI).

Why “the Uber of X” stopped working

When Uber was new, the analogy worked: people knew the concept, but it hadn’t been used for everything yet. After thousands of “Uber of” something, the expression became empty. It no longer says anything about your product.

Three simple positioning formulas

• [Familiar category] for [new audience/situation] · Example: Antivirus for apps made with AI
• [Result] in [time] without [pain] · Example: Contract reviewed in 5 minutes without a lawyer
• [Verb] + [object] + [for whom] · Example: Turns meetings into tasks for small teams

A phrase that works

• Fits in one breath.
• Uses words the customer uses.
• Shows the problem it solves.

A phrase that doesn’t work

• “Revolutionary,” “disruptive,” “AI-powered.”
• Technical jargon only you understand.
• Tries to say everything the product does.

The bar test

Say the phrase to someone outside your field. If they reply, “Oh, I get it,” and can repeat it to someone else, the phrase is ready. If they ask, “But what exactly does it do?” go back to the draft.

A short video is an elevator pitch

If it fits in 30 seconds, you know how to sell it

What it is

The idea that a product you can explain well in a short video already has the structure of a pitch for customers and investors.

Why learn this

Short formats make you cut the excess. What remains is the core value, which works in any sales conversation.

Key concepts

Elevator pitch, short video, core value, investor, demonstration.

An elevator pitch is the explanation you’d give someone during an elevator ride: 30 seconds or less. If your product works well in a short video, you already have that pitch. A short video is a pitch structure: hook, problem, demonstration, result.

The same worksheet works for

• A 20-second video on social media.
• The opening of a sales meeting.
• Your first words to an investor.
• The top of your home page.

Signs that the pitch doesn’t exist

• You start with the technology you used.
• You need a context slide first.
• The demonstration takes more than a minute to show anything.

Try it now

Generate 5 one-line positioning phrases and choose the best one using clear criteria. Paste them into Claude Code or the AI chat you use.

Você é um especialista em posicionamento de produtos digitais.

Meu produto: <descreva em 2-3 frases o que ele faz>
Público principal: <quem usa, com que profissão/situação>
A dor que resolve: <o problema, nas palavras do cliente>
O resultado que entrega: <o que a pessoa tem depois de usar>

Tarefa:
1. Escreva 5 frases de posicionamento de UMA linha (máximo 8 palavras cada).
   - Use pelo menos uma com a fórmula "[categoria conhecida] para [situação nova]".
   - Proibido: "revolucionário", "baseado em IA", "plataforma", "o Uber de".
2. Para cada frase, dê nota de 1 a 5 em: clareza para leigo, memorabilidade, diferenciação.
3. Escolha a melhor e explique em 2 frases por quê.
4. Diga qual pergunta um leigo ainda faria depois de ouvir a frase escolhida.

How to check: You get 5 short phrases with scores and one selected as the best. Read the selected phrase aloud to someone outside your field; if they can repeat it easily, it’s ready.

Practical tip

Keep the winning phrase somewhere visible. It will show up again in the page title, social media bio, opening of each video, and email subject lines.

Fewer clicks to value

Every extra click loses people

What it is

Map the user journey so the first useful result takes as few steps as possible.

Why learn this

When attention is short, every form, confirmation, or extra screen is a point where someone may give up.

Key concepts

User journey, time to value, friction, funnel, first result.

The user journey is the path someone takes from arriving at the product to experiencing its value. The most important measure along that path is time to value: how many clicks and seconds it takes for them to think, “This solves my problem.”

Terms in this topic

• User journey: The sequence of screens and actions between arriving and getting the result someone was looking for.
• Friction: Anything that takes effort without providing value: a long sign-up, email confirmation, or welcome screen.
• Funnel: A view of the journey as stages, showing how many people move from one to the next.
• Time to value: How long it takes from arrival for the user to see the first useful result.

How to cut clicks

1. Map the current journey. List every screen and click, from arrival to the first result. Count them.
2. Mark what doesn’t provide value. Sign-up, confirmation, welcome tour, plan selection. Ask whether each one can wait until after the result.
3. Move sign-up until later. Let the person try it first; ask them to create an account when they want to save or keep what they made.
4. Measure again. Compare how many people reach the result before and after the change.

Short journey

• Text box on the first screen.
• Result before sign-up.
• One main button per screen.

Long journey

• Five-screen tour before getting started.
• Credit card required to try it.
• Three equivalent options on the first screen.

Practical tip

Ask someone to use the product while you watch quietly. Each time they hesitate, note the screen. Those hesitations are your friction map.

Validate the interface with a free prototype

A UI that works for any niche

What it is

Launch a simple, free tool to test whether people in different audiences understand an interface format before using it in the main product.

Why learn this

It's an inexpensive way to find out whether your interface works without putting the paid product at risk.

Key concepts

Prototype, interface validation, diverse audience, proof of use, reuse.

A smart way to validate the main product's interface is to first launch a small, free tool with the same way of using it. If people from different countries, languages, and fields can use it without help, the format is validated.

Example: the page diagnostic tool

A creator published a free tool: people pasted the address of their web page and received a full assessment of how it was put together. It was one field, one button, and a report.

Weeks later, people were still finding the old post and saying thanks. Users with different languages, niches, and technical skill levels used it without asking anything. The conclusion: the “paste the link, get the diagnosis” format works for any audience. That format was reused in the paid product.

A good validation prototype

• The same way into and out of the main product.
• Free and no sign-up.
• Useful on its own, so people will share it.

A prototype that teaches you nothing

• A different format from the product you want to sell.
• Only friends testing it.
• No way to see how people used it.

Distribution bonus

A useful free tool is also a way to attract people: those who use it discover you. Discreetly add a link to the main product at the end of the result.

Prove value in less than two minutes

For people who aren't technical

What it is

The standard is that someone without technical knowledge should understand why the product is worthwhile, use it, and see the result in less than two minutes.

Why learn this

If only technical people can see the value, your market is small. If it takes more than two minutes, most people give up before then.

Key concepts

Nontechnical person, two minutes, paste-connect-done, one-click connection, interruption pattern.

The standard is simple: can someone outside the field understand why this is valuable, use it, and see the value in less than two minutes? The ideal action takes just a few steps: click, drop a file, paste a link, connect an account. And that's it.

Passes the two-minute test

• One-sentence explanation on the home screen.
• Ready-to-use example that you can try without your own data.
• One-click connection to the tools you already use.

Fails

• Requires a connection key between systems before first use.
• An empty screen waiting for you to figure out what to do.
• A result only an expert knows how to interpret.

How to stand out from copycat demos

Most AI product pages look the same: a big headline, a gradient, and three feature cards. One way to break the pattern is to do something different right away: for example, put the test field at the top of the page instead of a “get started” button that takes you to a sign-up form.

Test it with a real beginner

1. Choose someone outside the field. A relative, a neighbor, or a friend in another profession.
2. Give them only the address, with no instructions. Start a timer and stay quiet.
3. Stop at two minutes. Ask: “What does this do?” and “Would you use it?” Write down where they got stuck.

Practical tip

A preloaded example (“try it with this demo link”) solves half the problem. You see the value before you have to decide whether to trust me with your own data.

The content flywheel

Anonymized data that brings in traffic

What it is

A cycle where product use generates data, the data becomes useful public content, and the content brings in new users.

Why learn this

Increases your chances of getting discovered: people who have never heard of you find you through search, AI assistants, and social media.

Key concepts

Flywheel, anonymization, SEO, AEO, GEO, chances of getting discovered.

A flywheel (in English, flywheel) is a cycle that feeds itself: the more it spins, the easier it gets to spin. In the content flywheel, usage data becomes articles, articles bring in users, and users generate more data.

Terms in this topic

• Anonymize: Remove anything that identifies people or companies, leaving only the overall pattern.
• SEO: Optimizing to appear in search results, such as Google.
• AEO: Optimizing to be the answer cited by answer engines and assistants.
• GEO: Optimizing to be mentioned by generative AI (ChatGPT, Claude, Gemini) when someone asks about the topic.

Example with the security product

• The product finds flaws in thousands of apps built with AI.
• Every week, it automatically groups the most common flaws without identifying anyone.
• Publishes articles such as “The 5 vulnerabilities you need to fix in your AI-built app today.”
• People who are concerned and search for the topic find the article and discover the product without knowing who you are.

• Short videos. Short-video platforms don’t require followers, just consistency. A new account can have a video get 30,000 views and bring in customers without any promotion.
• Automatic articles. Regular reports generated from overall data and reviewed before publishing.
• AI answers. Clear content with numbers and lists is more likely to be cited when someone asks an assistant.

Privacy is not optional

Anonymizing is required and must be covered in the terms of use. Publish overall patterns (“37% of the apps analyzed had X”), never identifiable cases. A leak in your own content destroys the product, especially if the product is about security.

Module 3.2 summary · Positioning and journey

• One line: A familiar category + a new situation; no clichés.
• Short video: Hook, problem, demonstration, result: the same pitch worksheet.
• Fewer clicks: Show the result before sign-up; every extra screen loses people.
• Free prototype: Validates the interface format with different audiences.
• Two minutes: A nontechnical person needs to understand it, use it, and see its value in that time.
• Content flywheel: Anonymized data becomes articles that bring in users.

The feedback no one gave

Session recordings show where people get stuck

What it is

Tools that anonymously record how users browse your website or app, click by click.

Why learn this

Almost no one writes in to complain. Watching real usage reveals where people give up and why.

Key concepts

Session recording, PostHog, Hotjar, abandonment, production error.

Tools like PostHog and Hotjar offer session recordings (session replays): you watch a user navigate your product, like a video. Where they clicked, where they stopped, where they went back, and where they gave up.

Terms in this topic

• Session recording: A video reconstruction of a user’s browsing, based on page events. It doesn’t use the person’s camera.
• Abandonment: The moment a user leaves without completing what they came to do.
• Production error: A failure that only appears with real users in the published system, not in your tests.
• Masking: A feature that automatically hides sensitive fields (passwords, cards) in recordings.

A habit of big companies

In Stripe’s first few months, the team watched how its customers used the product, one by one, to find every pain point. This isn’t a habit for big companies; it’s what successful companies did at the beginning.

Many errors only show up this way: a button that fails when clicked twice, a form that disappears on a specific phone, a user trying something you never imagined.

What to look for in recordings

• Repeated clicks in the same spot (something didn’t respond).
• Scrolling up and down (the person is lost).
• Leaving right after a specific step.

Things to watch out for

• Recording sensitive fields without masking them.
• Not mentioning the data collection in your privacy policy.
• Watching one recording and assuming it applies to everyone.

Practical tip

Building your own AI session recording tool is possible, but established tools are still more reliable and have a free plan. Save your energy for your product.

The /product-feedback request

Recordings → video model → correction plan

What it is

An automated process in which an AI model watches session recordings, summarizes the problems, and another model turns that into a correction plan.

Why learn this

Watching recordings by hand doesn’t scale. With AI, you get a summary of dozens of sessions and a ready-to-use plan with a single action.

Key concepts

Skill, video model, Gemini Flash, Claude Code, correction plan, automation.

Today, instead of watching every recording yourself, a fast multimodal model (like the Flash versions of Gemini) can watch them all and tell you where the journey is confusing and where people don’t reach the desired click in less than ten seconds. Then, a coding agent like Claude Code turns that diagnosis into a correction plan.

Terms in this topic

• Skill: A saved set of instructions that Claude Code runs when you call it by name, such as /product-feedback.
• Multimodal model: An AI model that understands images and video as well as text.

Try it now

Create the /product-feedback skill in Claude Code. Save it as .claude/skills/product-feedback/SKILL.md in the project folder.

---
name: product-feedback
description: Analisa gravações de sessão recentes e gera um plano de correção da jornada do usuário.
---

# /product-feedback

Objetivo: descobrir onde os usuários travam, sem ninguém ter reclamado.

1. Baixe as <N, ex.: 30> gravações de sessão mais recentes do <PostHog | Hotjar>
   usando <a conexão entre sistemas/programa de linha de comando da ferramenta, com a chave em .env>. Pule sessões com menos de 10 s.
2. Para cada gravação, peça ao modelo de vídeo <ex.: Gemini Flash> um resumo com:
   - objetivo aparente do usuário;
   - onde hesitou (cliques repetidos, rolagem de busca, pausas longas);
   - se chegou ao resultado principal em menos de 10 segundos;
   - onde saiu.
3. Agrupe os problemas por tela e ordene pela quantidade de sessões afetadas.
4. Leia o código das telas mais problemáticas em <pasta do front-end> e escreva
   PLANO-FEEDBACK.md com: problema, evidência (quantas sessões), causa provável,
   mudança proposta, arquivo a alterar, como medir se melhorou.
5. NÃO altere código. Pare e me mostre o plano para aprovação.

How to check: Run /product-feedback in Claude Code. A PLANO-FEEDBACK.md should appear with problems ordered by number of sessions, and no code files should be changed.

Let the AI handle it

• Watch and summarize dozens of recordings.
• Group problems by screen.
• Suggest changes and point out files.

Keep this with you

• Approve what changes in the product.
• Decide what’s a priority for the business.
• Check that the recordings respect privacy.

Cost and privacy

Sending videos to an external model costs money per use and means sending browsing data outside your organization. Use recordings with masked fields, limit the number of sessions per round, and confirm that your terms of use allow this processing.

Copy the onboarding of someone who’s already making money

Inherit hundreds of A/B tests

What it is

Study, step by step, the onboarding of products that are proven to make money, and adapt what works to yours.

Why learn this

Companies that are already making money have tested the journey dozens or hundreds of times. Studying the final result lets you inherit that learning without paying for it.

Key concepts

Onboarding, reverse engineering, A/B test, verified revenue, computer use.

Onboarding is how a user starts using a product: from their first visit to the moment they understand it and start using it. Products that make money have refined their onboarding through many tests. Instead of starting from scratch, study what they do.

Terms in this topic

• Onboarding: The process of welcoming and guiding a new user through their first successful use.
• A/B test: Show two versions to different groups and see which one converts better.
• Reverse engineering: Study a finished product to understand how it was designed.
• Computer use: An AI agent’s ability to open a browser, click, scroll, and read the screen like a person.

How to study onboarding

1. Find out who’s really making money. Some websites list products with revenue verified through payment processor integrations. Revenue is a sign that the onboarding works.
2. Go through the entire onboarding. Create an account and try it for the first time. Or ask an agent with computer use (Codex has a built-in browser) to do this and record every screen.
3. Note what’s different. The order of the questions, what’s asked and when, the button text, what appears right after sign-up.
4. Adapt, don’t clone. Take the principle, not the design. What works for their audience may need adjusting for yours.

Why not just ask for “good onboarding”?

If you ask the model for onboarding without a reference, it gives you the average of what it saw in training. Average isn’t what converts best; it’s just what’s most common. Real references lift the result above average.

Practical tip

Be careful with public metrics platforms: showing your revenue on an open site also alerts copycats. Study others, but think carefully before sharing your numbers.

Design without bloating the workspace

Have the agent read references, not activate packages

What it is

Instead of activating several skills and built-in design features, ask the agent to browse reference galleries and handpick what fits your niche.

Why learn this

Different design skills bring styles that clash, bloat the context, and lead to the same generic look everyone has.

Key concepts

Design skills, conflicting design approaches, reference gallery, 21st.dev, whatships, Awwwards, handpick.

A common recommendation is "activate these seven design skills so your app doesn’t look AI-made." The result usually looks... AI-made. Worse: each skill brings its own style, and when the agent uses three at once, the design approaches clash. The workspace gets bloated, and the look gets confusing.

• 21st.dev. Gallery of interface components: backgrounds, borders, cards, buttons. Useful for seeing visual patterns.
• whatships. Product launch videos. A reference for your site’s marketing video.
• Awwwards. Award-winning sites of the month and year. Above-average typography, layout, and interaction.

Try it now

Ask an agent with a browser (Codex with computer use, or Claude with a browser) to study references and suggest a visual guide for your product.

Use o navegador para estudar referências de design. NÃO ative skills, recurso embutido nem pacotes.

Meu produto: <frase de posicionamento de uma linha>
Público: <quem usa>
Sensação desejada: <ex.: confiável e calmo | rápido e ousado>

1. Abra https://21st.dev e role pelos componentes (fundos, bordas, cartões, botões).
   Anote 5 que combinam com o produto, com o motivo.
2. Abra https://whatships.com, assista a 3 vídeos de lançamento de produtos parecidos
   e descreva quadro a quadro o que torna cada um claro em menos de 2 minutos.
3. Abra https://www.awwwards.com (sites do mês/ano), percorra 3 sites até o fim e anote:
   tipografia, posição dos botões, ritmo da rolagem e o que é incomum.
4. Escreva GUIA-VISUAL.md com: paleta, 2 fontes, 5 componentes escolhidos,
   estrutura da página inicial e um roteiro de vídeo de 60 s.
5. Justifique cada escolha pelo público e pela sensação desejada.

How to check: The agent should open all three sites (you can see the browsing) and deliver a GUIA-VISUAL.md with justified choices, without activating anything in the project.

Value with every scroll

As you review your page, ask at each scroll: "what did the person learn here?". Minimalist doesn’t mean empty; every scroll should deliver value.

Simulated users with subagents

Personas that test before the audience

What it is

Create AI agents that take on different user profiles, use or analyze the product, and give feedback from each person’s point of view.

Why learn this

It’s a quick, low-cost way to see the product through other eyes before exposing real people to it.

Key concepts

Subagent, persona, profile, simulated feedback, outside perspective.

A subagent is an AI agent that the main agent creates to handle a separate task, with its own instructions. You can create several, each with a persona: a user profile with a different age, routine, patience, and experience.

Terms in this topic

• Subagent: A helper agent created for a specific task, with its own context and instructions.
• Persona: A description of a typical user: who they are, what they want, and what annoys them.

Try it now

Run a panel of 4 simulated users in Claude Code on your page or app.

Crie 4 subagentes, um para cada persona abaixo. Cada um deve analisar
<URL da página | pasta do projeto> do próprio ponto de vista, de forma independente.

Personas:
1. <Nome>, 55 anos, <profissão>, usa o celular para quase tudo, desconfia de pagar online.
2. <Nome>, 20 anos, TDAH, testou 100 produtos na última semana, decide em 10 segundos.
3. <Nome>, gestor(a) que aprova a compra: quer custo, segurança e retorno claros.
4. <Nome>, especialista técnico cético: procura falhas e promessas exageradas.

Cada subagente responde:
- Em 10 segundos, entendi o que é? (sim/não + o que achei que era)
- O que me faria sair da página.
- O que me faria pagar.
- A frase exata que eu mudaria.

Depois, consolide em PAINEL-PERSONAS.md: pontos em que 3 ou mais concordam,
divergências importantes e as 5 mudanças de maior impacto, em ordem.

How to check: PAINEL-PERSONAS.md should appear with four distinct opinions and a prioritized list. If all four sound alike, make the personas more specific and run it again.

Simulation is not validation

Simulated personas help you find obvious problems before launch. They don’t replace real people paying. Use them to improve the product, not to conclude that it will succeed.

Synthetic crowds and judgment

Thousands of simulated buyers, and what stays with you

What it is

Tools that simulate hundreds or thousands of buyers, each with a short profile, to predict the market’s reaction; and the course conclusion.

Why learn this

They let you test messages and pages at scale before spending money; and they remind you that when anything can be built, judgment is what sets you apart.

Key concepts

Synthetic crowd, Mirofish, backtest, system prompt, judgment.

There are tools, such as Mirofish, that create hundreds or thousands of agents, each with a short system prompt describing who they are. Together, they simulate buyers in a market and say what they would think of your product, page, or ad.

Terms in this topic

• Synthetic crowd: A large number of agents with varied profiles that simulate how an audience might react.
• Backtest: Testing a method with past data to see whether it would have correctly predicted what actually happened.
• System prompt: An initial instruction that defines who the agent is and how it should behave.

The test that builds confidence: the backtest

People have been using these tools for a simple test: take the last 10 posts from an account and ask a synthetic crowd matching that account’s audience to say which one would be most successful. In many cases, it points to the post that actually went viral.

If the simulation gets the past right, you gain some confidence in using it for future decisions: which title to test, which page to publish, which offer to make.

Use simulation to

• Choose between title or page versions.
• Find objections before spending money on ads.
• See how different audiences react.

Don’t use simulation to

• Conclude that the product will sell.
• Skip conversations with real customers.
• Justify a decision you had already made.

Conclusion: judgment is the game

Everything this course has shown is within anyone’s reach today: building, designing, simulating, analyzing. That’s exactly why none of it sets you apart on its own.

What sets you apart is judgment: knowing whether something is worth building, for whom, for how long, and how to reach those people. In a world where anything can be built, that judgment is the whole game.

Your next practical step

Take the product or idea you’re working on and go through these in order: the one-line statement (3-2), the persona panel (3-3), and the plan to reach the first 100 people (3-1). In one afternoon, you’ll have all three answers written down.

Module 3.3 summary · Feedback, design, and simulated users

• Session recordings: Real feedback is in what the user did, not what they said.
• /product-feedback: A video model summarizes the sessions; a coding agent puts together the plan; you approve it.
• Onboarding into an established product: Study who is already making money; adapt the principle, don’t copy the design.
• Design by reference: Browse galleries and choose carefully instead of piling on skills.
• Personas and crowds: Simulation finds problems early, but it doesn’t replace a paying customer.
• Judgment: When anything can be built, knowing what’s worth building is the game.