Building Product Moats Against AI
Last week, I spoke with an advisor who looks at software companies all day, every day. I was the seventh founder he had talked to that day, and every single conversation he has is the same these days.
It essentially boils down to this:
What are you going to do to build a moat against AI?
If you were to ask ChatGPT or Claude whether your business could be replaced by AI, what would it say? What would it say you could do to build a moat around your business against AI?
AI is the most formidable competitor software companies have ever faced.
However, it is still another competitor.
With our company, Interact, we have faced many waves of competition. They started with competitors that had small amounts of funding and scaled up to companies raising absolutely enormous rounds designed to knock us out of the race.
Every single time, we have found a way to survive and grow.
I will be blunt. AI is the most formidable competitor that has ever come up. It is the scariest and the most potentially debilitating of any competitor we have faced.
That is true not just for our company, but for pretty much every software company.
However, at this point, there are some very clear things that AI is not going to be able to execute well on. At least not anytime soon.
These are the four product moats against AI that I see.
1. Product Experience and Taste
The first product moat against AI is what is being called “taste.”
I think taste is often used as a cop-out term. What it really means is this:
Does your product feel like it is easy to use?
Does it feel like it does the thing it is supposed to do?
Does it give the user or buyer the satisfaction they want from using your product?
There are a million things that go into creating that feeling. But after building products for the last 15 years, I believe it really boils down to understanding what people want to do with your product, then making it easy and enjoyable for them to do that thing.
That is much more difficult than it sounds.
You have to understand why people are using the product in the first place. You have to know what outcome they are trying to reach. Then you have to remove everything that gets in the way of reaching that outcome.
If you can accomplish that, I think you are totally set.
That is one huge moat against AI.
2. Visual Design
The second moat is similar to the first, but it is specifically about visual design.
You may have noticed that nearly everything built purely with AI has a very similar design.
This is ironic because AI was supposed to be the thing that helped your unique vision come to life. Instead, it has turned out to be the thing that makes everything look the same.
This is not particularly surprising.
There is technically a most efficient way to design products, and that is what AI is going to default to. It will choose common layouts, familiar components and patterns that appear most frequently in its training data.
The result might be functional, but it rarely feels distinctive.
If you can build a design system that makes people feel like your product is elevated, unique and different, you will have a moat.
People will feel something when they use it. They will recognize it. The product will have an identity that cannot be reproduced simply by asking an AI to build another version of the same software.
3. Integrations
The third moat is integrations.
Early in the current AI cycle, there was a moment when everyone thought integrations might not matter anymore.
Maybe AI would move data between systems automatically. Maybe direct integrations would become unnecessary. Maybe every product would simply communicate with every other product through an AI layer.
Now that we have gone a little further into the cycle, it is clear that this has not happened.
A direct integration through an API, built into your product and made very easy to set up, is still a key differentiator.
Customers do not want to manually move information around. They do not want to build and maintain complicated workflows. They want to connect the tools they already use and trust that everything will continue working.
The deeper and more dependable those integrations become, the more embedded your product becomes in the customer’s business.
That creates a real moat.
4. Customer Support That Solves the Last-Mile Problem
The fourth moat is customer support that helps people get where they want to go.
This is also emerging more clearly at this stage of AI.
Ironically, you are starting to see AI companies doing much more customer support to help people use AI.
The entire point was supposed to be that AI would help you use AI. But it turns out there is still a last-mile problem.
The technology might be capable of doing something, but the customer still has to figure out how to apply it to their particular situation. They need to configure it, connect it to the rest of their business and get through the inevitable moments when something does not work as expected.
AI can provide answers, but it does not always understand the full context of what the person is trying to accomplish.
If you can solve that last-mile problem using trained people who can walk customers through the process and actually solve the issue, that is a moat.
The goal is not simply to respond to support tickets. It is to help customers reach the outcome they bought the product to achieve.
Using AI to Strengthen the Moat
These are the four moats I have seen:
A product experience that makes it easy and enjoyable for people to accomplish what they came to do
A distinctive visual design system
Deep, reliable and easy-to-use integrations
Customer support that solves the last-mile problem
If you combine these with an AI-driven product development process, you can move extremely fast while continuing to strengthen your core business.
You can take advantage of the increased speed AI offers without allowing your product to become another interchangeable piece of AI-generated software.
You will be able to move faster, do more and expand into new markets. At the same time, you will be building something that remains defensible based on where AI is today.
Of course, this is all subject to change. There is always the possibility that AI will expand further into these areas.
But we are almost four years into this AI cycle, and these four areas have emerged as predictable things that AI does not execute particularly well.
If I were a betting man, which I am, these are the four areas I would bet on.
