📊 Building with an LLM means building the average
In simplified terms, a language model works by predicting the most likely continuation of a text. It learned to do this by reading a huge amount of existing content. The natural result is that, without strong guidance, it returns what's most common: the most common layout, the most common marketing line, the most common user journey.
⚖️ The average isn't a flaw; it's the nature of the tool
For tasks where the common option is the right one (a login form, a confirmation email), the average is exactly what you want. The problem appears when you use the same tool, in the same way, for the part of the product that should be unique.
💡 Where to accept the 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
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 line, a product that does something you didn't know was possible. In marketing, this is called a pattern interrupt.
The average problem 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 to summarize meetings impressed people; today, it's expected.
✓ Interrupts
- ✓A demo that solves something in seconds, right in front of someone.
- ✓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
We're entering a world where few people will exercise their own judgment. Everything can be outsourced: the copy, the design, the strategy, even the user journey—the sequence of steps someone takes from arriving at your product to getting what they came for.
🛡️ 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 is valuable in an inflated market. Being careful, thinking about the user's experience, questioning the model's suggestion: all of this becomes a competitive advantage.
💡 A one-week exercise
Before asking AI for 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
Open any tech platform and count how many times the word "launching" or "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.
📊 The typical response to excess
- •Someone scrolls for 40 minutes and saves six things “to watch later.”
- •The next day, six more appear. Yesterday’s are never opened again.
- •Bookmarks become a graveyard of good intentions.
⏱️ What this demands from 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-generated,” and the result still looks AI-generated. Following the trendy formula means, by definition, producing the trend’s average.
🗺️ A crowded market isn’t a closed market
If everything can be copied and no one has a moat, does that mean no product deserves to exist? No. When you look at a problem with enough precision, you almost always find parts no one has solved, or that have been solved the wrong way, or that have been solved but never reached the people who need them.
✓ Questions that reveal gaps
- ✓Who is handling this with a spreadsheet or on paper today?
- ✓Which existing solutions 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?” (It almost always does.)
- ✗“Is this innovative?” (Innovation isn’t a requirement.)
- ✗“Can you do this with AI?” (You can do almost everything.)
💡 Distribution can be a gap too
Often, the solution exists but has never reached the right audience. Bringing a known solution to an underserved niche, in its language and in a way that works for it, is a legitimate opportunity.
🔭 The three forms of judgment
Building in a time when you can build almost anything takes 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.
| Level | Question | Who already does it well |
|---|---|---|
| Building | How do I make this work? | Anyone with a coding assistant |
| What to build | Which problem should I solve? | Someone who knows a niche well |
| Whether it’s worth building | Does this deserve to exist a year from now? | Few people. That’s what this course trains. |
How to use all three in practice
Start with the third
Before any prototype, answer: is it worth building? Track 2 is all about that question.
Then the second
If it’s worth building, what exact part of the problem will you tackle first?
Only then the first
Now you can build quickly, with all the help AI offers.
💡 A phrase to stick on your monitor
In a world where you can build anything, judging what’s worth building is the whole game.
🧪 Quick module quiz
Three questions. Click an option to see the answer.
1. Why does a product built only with an LLM’s default suggestions tend to feel generic?
2. What happens to a pattern interrupt when everyone copies it?
3. Which of the three kinds of judgment became almost free with AI?
📋 Module summary
Next module:
1.3 - Productive Pessimism