You can clone the product (most probably), but you can't clone the company
What a Strategy/BizOps Mind Sees When Building a Company: #7
Chatting with someone this week, and I learned that Bain is now using vibecoding, getting AI to replicate a target’s software, to pressure-test the competitive advantage of acquisition targets. If the model can rebuild your product over a weekend, the logic goes, you don’t have much of a moat. These replicas are apparently already shaping specific purchasing decisions.
I agree this should be a filter. But it’s a first-pass filter, not a verdict.
Vibecoding is a first filter not a verdict
Here’s what it actually tests: whether your features are defensible. And increasingly, for almost everyone, the answer is no. That isn’t news. It’s the whole reason I keep saying you have to build something the model can’t eat. If your product is a thin layer over capability that commoditises by the month, vibecoding just makes that obvious faster.
But a company is much more than a collection of features.
I wrote a while back about how AI changes the value chain of a company.
The short version: everything a company produces runs through a flow of data → application → implementation. A feature isn’t a step in that flow. It’s the output of running the whole thing. Whether you “create a product feature, or a report for the board,” you start with data, apply it to a problem, and ship the result.
Which is exactly why vibecoding is a narrow test. It clones the technical output: the finished software, as it appears on the surface. It does not clone the chain that produced it: the proprietary data that went in, the judgement applied to it, or the trust and distribution wrapped around it. Cloning what a company shipped is not the same as cloning how, and why, it can keep shipping.
Cloning what a company shipped is not the same as
cloning how, and why, it can keep shipping.
So no, it doesn’t eat the audience you’ve earned, the customers who stay, the distribution you’ve built, or the way your team actually innovates. Those are the things that compound, and none of them show up when a model clones your UI.
I have spoken about how companies are revenue generating systems. And more than ever we see that shipping a product is only a small part of that system. The much larger part is the kernel of expertise and judgement that connects an interested ICP and a vision for a solution.
Final Word
So by all means, run the vibecoding test. If a target’s product can be trivially regenerated, that’s a real signal about pricing power, about how much of the value genuinely sits in the software. But treat it as the cheap, fast screen it is. The expensive question, the one that actually determines whether an acquisition is worth it, is everything the model can’t replicate: who the company reaches, why those people stay, and how it keeps making good decisions once the easy stuff is automated away.
Replicating software has never been the hard part. Building a company around it is.