Four Things AI Can't Steal from You

A while back, a founder came to me for tea. He looked wiped out.

He said his team had just shipped a new feature on Monday — built on the latest model, with stunning results. And guess what? By Friday, a competitor's product had something almost identical.

I asked him, "So what are you going to do?"

He gave a wry smile. "Just build the next thing. Whatever we ship gets lifted over the weekend anyway."

That conversation stuck with me for a long time.

This generation of models has raised everyone's floor. The problem is, it has also leveled the ceiling — everyone now reaches roughly the same height. You call an API today; your rival calls the same API tomorrow. You think you're building rockets, but everyone's buying parts from the same shop.

Which leads to a genuinely dangerous question. When the underlying capability is something anyone can buy, what exactly do you win on?

Let me throw cold water on this first

Most people's first instinct is: I'll just run faster, get bigger.

Is scale alone enough to keep you safe?

I saw a set of numbers that hit hard. Companies flying the AI flag have a median net revenue retention of just 48%. What does that mean? Of the money they earn this year, more than half leaks away by next. For comparison, traditional B2B software companies sit at 82%.

Growing like a weed on one side, leaking like a sieve on the other. Getting bigger is not the same as getting stable.

So the question you really should be asking is the exact inverse of "what do I have in my hand?" Swap it for this one:

What would my customers lose if they left me?

Think that question through, and the moat（a durable competitive advantage rivals can't easily cross）problem solves itself. Put plainly, a moat is the cost someone pays to lift what you've built.

I break it into four things. These four, the model can't walk off with, and competitors can't copy fast enough.

The first: you gave the problem a name

Let me ask you first — what does it mean to "own a category（a recognized class of product or solution in buyers' minds）"?

A lot of people think it means carpet-bombing the market with ads, or holding the number-one market share. Neither. It means that the moment a need pops into a user's head, the first word that jumps to mind is your name.

Reach that point, and your rivals are immediately on the back foot. They have to spend energy explaining "I'm not that other guy" before they can even start selling their own thing. You're already standing on the finish line; they're still queuing behind the starting line.

There's a company in voice AI that plays this harder than anyone. Among first-time buyers of this kind of product, nine and a half out of ten come in through their door. Why? Because they're not just a useful tool — they've become nearly synonymous with the category. You say "voice AI," and users reflexively think of them.

That's real positional sense.

You might say, those guys have money, data, a first-mover advantage — how's a small company supposed to grab that?

Think about it: does naming a new problem have anything to do with company size? Big companies often can't be bothered to name a new need, because their existing business is still minting money. That leaves a gap — and as a latecomer, you can step in and define a track the incumbents never even paid attention to.

Give it a name. Articulate a vague pain clearly. Make the market start thinking in your language.

That's the first thing they can't lift from you.

The second: the more it's used, the stronger you get

Scale on its own isn't scary. What's scary is when scale makes something better and better to use.

What do I mean?

Take this example. There's a company in enterprise spend management that has served more than 70,000 customers, with over 200 billion in annual procurement flowing through their hands. That sounds like a big number, but the real trick isn't in the number itself.

From that 200 billion, they see what most people can't: which suppliers are quoting high, which departments' procurement looks off, how much peers pay to land the same item. The more customers use them, the better they understand the market; the better they understand the market, the more money they save customers.

And there you have a closed loop. Every new customer, on the surface, is one more source of revenue. Underneath, it's another dose of data fed into the whole system. The system gets smarter, old customers get more locked in, and new customers get more eager to come.

That kind of scale is alive.

Flip it around: if adding customers just means you sell a few more copies of the identical product, that scale is dead. A competitor with money can pile it up the same way. For scale to become a barrier, the act of "using" itself has to become the fuel that evolves the product.

The third: replacing you means tearing it all down

This one's the most ruthless.

What does real embedding（here, deep integration into a customer's workflow — not the machine-learning sense）mean? It means that after a customer has used your product long enough, their workflow, their organizational structure, even their headcount is designed around you.

At that point, replacing you isn't as simple as buying a substitute. They'd have to rip out the whole way they operate and start over.

I once saw a company that makes critical equipment for data centers. They don't just sell machines. They remotely monitor the equipment while it runs, predict failures before they happen, dispatch someone the moment something goes wrong, and back it all with a full ongoing service.

Customers keep using them, and little by little their own operations team gets leaner — some roles stop getting filled entirely, because all of that has been outsourced.

Guess what: can those customers still switch suppliers?

Switching means rebuilding a whole monitoring system, rehiring a crew of people who actually know the gear, and living through the painful break-in period all over again. This isn't swapping a part. It's tearing down half the workshop and rebuilding it from scratch.

A good product can be substituted. What's grown into the bone can't.

The fourth: you know "why"

The last one is the most valuable — and the easiest to overlook.

Lots of data does not equal understanding the customer. A dashboard tells you what happened; only someone who truly knows the trade understands why it happened.

There's a company that does healthcare data. They hold molecular-level information — genes, pathology, all kinds of test indicators — tied together with patients' long-term treatment records.

That data, in itself, is of course valuable. But what makes them genuinely irreplaceable is what they can read out of that pile: which drug works better for which kind of patient, which indicator can foreshadow a change in condition, how the next clinical trial should be designed.

Other companies can also hoard data. But the ones who can extract the "why" from it? Vanishingly rare.

Scale amasses data; the "why" is the gold inside it.

These four don't line up in a row

At this point you might ask: so do I have to hold all four?

No.

A small company just starting out has little data, few customers, and nothing you'd call embedding. At that stage, where's the most honest place to begin?

With the first one — giving the problem a name. It's the only one that doesn't depend on scale or time, only on a sharp eye. Spot a pain nobody's defined, say it clearly first, and the market slowly starts thinking in your framing.

Once customers accumulate and scale starts feeding back into the product, the second one grows in. As customers use you more deeply and their workflow begins to revolve around you, the third takes root. And once you've read, transaction by transaction, why customers buy and why they stay, the fourth locks into place.

At its most mature, the four mesh together like a gear train. Naming drives down your customer acquisition cost（CAC）and speeds up scale; scale opens more use cases and embeds the system deeper; embedding generates exclusive data that makes you understand customers better; and understanding customers better, in turn, makes your naming sharper and more ruthless.

A rival can copy one of your moves. What they can't copy is the accumulated turning of this gear train over years.

One last line

In this era, anything that can be copied gets copied by the weekend.

Go build the four things that can't be lifted.