In Part 1, I looked at five popular stories about making money with AI.
Here are five more.
Same rule as before:
The stories aren’t necessarily false. The interesting part is usually what they leave out.
Myth #6: “There are thousands of free projects on GitHub. Just package one and sell it.”
This is actually more legitimate than many AI side-hustle ideas.
You don’t need to build a CRM from scratch if someone has already built a perfectly good one on GitHub.
Your value might be deployment, configuration, localization, a better interface, training, maintenance, or support.
In other words, you sell the last mile.
That can absolutely be a business.
But then comes the awkward question:
What happens after installation?
Who handles upgrades?
Who backs up the data?
Who fixes the server when it goes down?
What happens if the upstream project stops being maintained?
What if a customer accidentally deletes half the database?
And “free on GitHub” does not mean “free to do absolutely anything with.” Open-source licenses come with different obligations around distribution, attribution, and source code.
So yes:
“I didn’t write a single line of code” can be completely true.
But:
“I don’t need technical ability” usually isn’t.
The moment you start charging customers, you stop being someone who merely packaged some code.
You become the person responsible for making it work.
And that is often what the customer is actually paying for.
Myth #7: “AI can replace expensive consulting and market research.”
This is one of the most attractive AI business pitches:
“Traditional research takes weeks and costs tens of thousands of dollars. My AI personas can simulate 100 consumers in 30 minutes.”
Sounds fantastic.
One small question:
Have any of those consumers ever consumed anything?
Synthetic respondents can be genuinely useful.
They can help generate hypotheses, explore possible reactions, identify questions worth asking, and speed up the early stages of research.
The dangerous part is when:
“simulated consumers”
quietly becomes:
“consumers.”
Those are not the same thing.
If a company is preparing to make a major product investment based on your research, having at least a few participants who possess the inconvenient property of actually existing is probably still a good idea.
So the better proposition is not:
“AI replaces consumer research.”
It is:
“AI compresses the exploratory phase from three weeks to one day, so we know which hypotheses are worth validating with real people.”
AI may be most valuable not when it replaces professional judgment, but when it makes professional judgment faster.
Myth #8: “Digital products have zero marginal cost, so they are the perfect AI business.”
Prompt packs.
Ebooks.
Children’s books.
Notion templates.
Courses.
Knowledge bases.
Wallpapers.
PDF guides.
The economics look beautiful:
Create once. Sell forever.
And AI has pushed the cost of “create once” dramatically lower.
There is just one small problem.
If you can make it in two hours, why can’t everyone else?
When your production cost approaches zero, your competitors’ production cost approaches zero too.
Your AI-generated children’s book may have taken two hours.
Great.
But why should someone buy yours instead of the other 300,000 AI-generated children’s books?
That is the other side of zero marginal cost.
It applies to everyone.
So the valuable part of a digital product is increasingly not:
“I can generate it.”
It is:
“I know what to create, who wants it, how to reach them, and why they would choose mine.”
Production is becoming abundant.
Demand, distribution, taste, trust, and attention are not.
Myth #9: “Knowing how to use AI is becoming an increasingly valuable skill.”
This is half true.
Knowing how to prompt well, connect models, build workflows, and use AI tools can still give you an advantage today.
But ask a slightly uncomfortable question:
What if everyone knows how to do this six months from now?
“I use AI to write customer-service replies” is not much of a moat.
But this is different:
“I spent five years running ecommerce support. I know which complaints need immediate escalation, when a refund prevents a bigger loss, which customers are worth retaining, and which issues can hurt marketplace rankings. Now I’ve encoded that experience into an AI workflow.”
The first person is selling AI skills.
The second is selling:
domain expertise amplified by AI.
That distinction matters.
AI can amplify five years of experience.
It cannot retroactively give you the five years.
Myth #10: “Find a price discrepancy, let a bot trade it automatically, and enjoy effortless arbitrage.”
Few ideas combine AI, money, automation, and the possibility of getting rich while asleep quite as efficiently as this one.
Take prediction markets like Polymarket.
Build a bot.
Scan prices 24/7.
Compare them with news, betting odds, and external markets.
Spot mispricing.
Trade automatically.
Profit.
There may indeed be temporary price discrepancies and information delays.
But there is an obvious question:
If the arbitrage is that easy to detect, why would only your bot see it?
If an ordinary person can ask ChatGPT to help code the strategy, professional traders can do the same.
At that point, the game stops being about knowing the formula.
It becomes about:
better data,
lower latency,
faster execution,
more capital,
lower costs,
and stronger risk controls.
That juicy 1% “risk-free” spread on your screen may disappear somewhere between your program receiving the data and your order actually reaching the market.
So:
Finding arbitrage is not the same as capturing arbitrage.
And markets have a particularly efficient habit of making obvious, easily copied profits disappear very quickly.
The Bigger Pattern: AI Moves the Moat
Put these examples together with the five from Part 1, and a much more interesting pattern appears:
Every time AI lowers a barrier, it also tends to destroy some of the competitive advantage associated with that barrier.
When everyone can build a website, “I can build websites” becomes less valuable.
When everyone can produce 30 pieces of content a day, production itself becomes less valuable.
When everyone can generate prompts, ebooks, and images, the ability to generate them becomes less valuable.
When everyone can build an agent, reliability becomes more valuable.
When everyone can build a trading bot, data, speed, capital, and risk management become more valuable.
So AI does not simply turn:
“hard” into “easy,” and therefore “difficult to make money” into “easy to make money.”
Something more subtle happens.
When one difficult thing becomes easy, the competitive advantage moves to something else that is still hard to copy.
Code → demand and distribution.
Content → perspective and attention.
Agents → reliability.
Information → access to proprietary or continuously refreshed information.
Open source → service and accountability.
AI research → professional judgment.
Digital products → brand and channels.
AI skills → domain expertise.
Trading algorithms → infrastructure, capital, and execution.
The moat doesn’t disappear.
It moves.
So What Has AI Actually Changed?
There is a simple test I increasingly like for evaluating an “AI business.”
Remove the word AI.
Then ask:
Did this demand already exist?
Were people already paying for it?
What exactly does AI make better?
Resume writing was already a business.
Web development was already a business.
Consulting, consumer research, content production, software deployment, digital products, advertising, and trading all existed long before generative AI.
What AI has dramatically changed is their:
cost structure and production efficiency.
Something that used to take two hours may now take ten minutes.
Something that used to require a five-person team may now require one person.
A customer segment that was previously too expensive to serve may suddenly become profitable.
Those are real opportunities.
But there is an important distinction.
AI makes it easier to build a product.
It does not automatically find customers.
AI makes it easier to produce content.
It does not create more human attention.
AI makes it easier to write a trading program.
It does not magically create more arbitrage.
AI can generate a consulting report in minutes.
It does not absorb the consequences when the recommendation is wrong.
In other words:
AI is extremely good at changing how things get done.
So far, it has been much less effective at answering the oldest questions in business:
What should we build?
Who actually wants it?
Why would they pay us?
So my current formula for AI monetization looks something like this:
AI Monetization = Real Demand × Distribution × Judgment × Sustainable Delivery × AI Leverage
Most AI-money stories spend almost all their time talking about the final variable:
AI leverage.
Then they quietly assume everything before it equals 1.
Unfortunately, multiplication is unforgiving.
If any one of those variables is close to zero, the final result is still pretty close to zero.
So the next time you see:
“Launch in one day.”
“Fully automated.”
“$10K a month.”
“No experience required.”
“No customer acquisition needed.”
“Passive income.”
“Infinitely scalable.”
You don’t need to dismiss it.
There may genuinely be an opportunity there.
Just ask a few less glamorous questions:
Who pays? Why do they pay? Where do the customers come from? Who is responsible when delivery fails? What are the real costs? Are there copyright, platform, licensing, or compliance risks? And if everyone has access to the same AI tomorrow, what advantage remains?
If those questions have convincing answers, the opportunity is worth investigating.
If someone talks for two hours about prompts, agents, workflows, automation, and content generation but never seriously explains where the customers come from…
you probably haven’t found a business yet.
You’ve found a very impressive production tool.
Perhaps the biggest illusion of the AI era is that, for the first time, we have something approaching unlimited production capacity — and it is tempting to assume that unlimited production capacity also means unlimited earning capacity.
It doesn’t.
If anything, the opposite may be true.
As making things becomes cheaper, deciding what to make, who to sell it to, and why anyone should choose you becomes more valuable.
AI is changing a lot.
The basic logic of business?
Not nearly as much as we think.