“Building great AI products is like building any other great product.”
That’s what I hear around, and also what I thought for quite some time. Until I realized that it’s not true.

The basics are the same, yes. But we’re facing a paradigm shift, and applying the old playbook needs a revamp. Especially when it comes to monetization and scaling.

If you’re raising an eyebrow right now, hear me out. Here’s why I think building (and scaling) AI products is different.

What doesn’t change…

I am a firm believer that the non-negotiables still apply:

Know your users—start with their problems, not your tech, your idea, or your TAM.
Validate that the problem is big enough—for your users and for your business.
Understand your unit economics—how you’ll actually make money.

AI only makes these basics more important. The pressure to ship fast and cheap is high. But if you skip validation, you’ll end up as one of the thousands of AI products solving the same thing… and getting ignored.

Lean startup is not dead. Validation is not dead. If anything—they matter more than ever.

But, I also believe there are other aspects that are coming into play when building AI products, ones for which the playbook is different, and ones we should be reflecting on. Especially as there is a clear push to be building those AI products.

…What is changing

Three big shifts stand out to me, areas where AI is bringing its twist to the game:

Which problems you choose to solve.
The cost of scaling.
Where your moat really comes from.

1- Solving your customer’s biggest problem might not pay off

There is one counter-intuitive truth to building AI products: solving your “biggest” customer problem can be a trap.

Why? Because if the big players put their eyes on it, they will likely solve it faster, better, and cheaper than you. And they control the underlying tech and distribution, making the cost of switch to them extremely easy.

The AI twist to how we prioritize: resist the urge of solving your biggest customer problem and focus instead on anticipating where the technology is heading, before acting. This will put you one step ahead, avoiding to waste your time, sweat, and money to then see your hard created differentiation disappear overnight.

Example: Granola. Their CEO explained that they avoid problems they expect the big guys to solve next, even if they are the most pressing ones for their customers. Focusing instead on features that delight users.

Example: Nano Banana. Google’s latest (if you ask me mind-blowing) Gemini photo editor made many AI photo-editing startups irrelevant overnight.

What do these example teach us? You might have an idea, start in an opportunistic way, but if you picked the wrong bet, your differentiator, or even your product will be zeroed from one day to the next. You are using someone else’s technology, remember they can flip the table under you.

In the old product world, solving the biggest pain point was a recipe for success. In the AI world, this is not true anymore.

And as the game shifts, you need to be getting better at strategy, the one of the market, as well as your own. Building successfully means anticipating where the big guys are heading.

2- Scaling is different when every user costs you money

The old dream:
build a product > find market fit > scale at near-zero marginal cost.

Isn’t that beautiful? Thousands of users, coming for free.

But wait… AI doesn’t scale that way.

Every user costs. Every interaction costs. And going “full scale ahead” might be costing you more than what is making you.

The old playbook doesn’t work anymore. Not even the big guys have cracked the nut, and while we can speculate that they will turn on the distribution tax soon, how would you make money when your product depend on their tech and distribution?

Building AI means unlearning how you scaled products before, recalibrate your users monetization, and your ability to do that at scale (if you even should go there).

The implications:

You need ROI on much smaller numbers.
Churn hurts more—so loyalty and delight matter most.
Forget paid acquisition—growth comes from word of mouth. (think about it: how did you first hear about chatGPT, cursor, lovable, perplexity? Exactly..)
Business models need to optimize for loyalty and deep integration, not cheap volume.

The new growth playbook:
Solve a problem for a small set of users > delight them > leverage word of mouth > integrate deeply > prove so much value they’re happy to pay

You might have to rethink your scale ambitions. Because you might be able to build much cheaper, but scaling will cost more. Be honest: is this really how you would have thought about scale before AI?

3- Your moat won’t come from expanding your footprint

This isn’t just a break in the playbook caused by AI, but a shift in how growth strategies might play out—especially with the rise of hyper-personalization enabled by AI.

For years, we’ve seen companies chase the super app dream. From WeChat to Facebook, the idea was: expand your footprint, keep users inside your ecosystem, and capture more of their time and attention. Add more features, become “the destination.”

I made the case previously on why, based on my experience, this is rarely a successful strategy. Stretching too far beyond your core job-to-be-done often weakens your proposition

But now insert AI into the picture, and the cracks in this strategy become even more visible.

Why? Because AI makes it easier than ever to create or tweak products for yourself. I keep seeing tiny, personal variations of existing apps, built to scratch your own very specific itch.

Let’s take period tracking apps as a reflection point:

There are plenty on the market already, yet I’ve seen people build their own.
The difference often comes down to a small personal twist—something “perfect for me.”
That pride in building (and owning) your own app, combined with low cost of switching might get you into trouble. If you are developing a period tracking app, you might be losing your user base.

The signal here: hyper-personalization is rising, and what you once dismissed as “edge cases” may soon be your users’ baseline expectations. If you do not meet them, they might build it themselves, or go somewhere else where they get exactly what you need.

I believe this has two main consequences:

If your product is established, double down on your core. Make sure you’re serving your users’ real needs before you get distracted with shiny AI experiments. Otherwise, you risk losing them.
Rethink your moat. Maybe it’s community, data ownership, integrations, or unique experiences. But it won’t come from “just adding more features.”

The shift is subtle but important. Your users may not want more from you—they may want exactly what fits them best. And if you don’t provide it, they’ll either build it themselves or go somewhere else.

If you are busy testing AI to solve problems that are not important, maybe pushing the monetization of your core product to finance all of this, and do not have an high barrier to exit…I’ll let you finish the sentence.

Your core was always important, but now it is a bit more important. Or you might end up with a shiny AI and a lost market-fit.

To sum it up

Building with AI means the basics still matter, but the rules of the game have shifted.

If your product is new, recalibrate which problems to solve and monetize early.
If your product is established, protect your core, shore up your moat, and don’t let AI distraction cost you market fit.

The new paradigms of building AI products