1/ On moats barriers to entry
TL;DR Moats still exist
There has been a lot of talk about how moats are dead. This talk is completely misguided because switching costs and moats aren’t the same as barriers to entry, and AI doesn’t change it.
Let’s start with a basic example—your phone. How easy is it to switch from iPhone to Android?
It’s not that hard, but it is not that easy. If you are an Apple user and you buy a new phone, Apple makes it seamless to move all your apps, your settings, and data. For the average user, this is hundreds if not thousands of data points and settings.
You of course can redownload all the apps you use, learn and adjust new settings, and configure your Android. But that’s work and inertia, and Apple would need to really piss you off or Android needs to be substantially cheaper for you to leave.
Also recall that Apple envelops you into their ecosystem. Phone, AirPods, laptop, iPad, TV, etc., all orchestrate together, making switching costs high and making it even harder for you to leave.
Next, let’s look at Instinct vs. Muse—new personal agents that just launched, one from a startup and another one from Meta.
Again, don’t confuse fast follow with a high switching cost moat based on data captured.
It seems like Instinct has no moat, but that’s only because it is new to the market, hasn’t gained a real user base yet, and most importantly hasn’t accumulated enough vested users.
Let’s say you’ve been using an agent for quite some time and configured dozens or even hundreds of tasks for this agent. You shared your information, logins, preferences, and much more.
Similar to how switching from iPhone to Android isn’t easy, moving your data from one personal agent to another may not be easy either.
As of today, there are no compatibility regulations that force one agent platform to export data in a format that’s understood by all others, so once you go deep, you are kind of stuck.
And what makes you stuck is your own data, the very time you spent setting things up and customizing them.
Now, of course, if something remarkably better or much cheaper comes up, then you switch, but if things stay roughly the same, you don’t.
What’s true is in the age of AI, and especially consumer products, there is insane pricing and feature pressure, but that’s not the same as data moats.
I deliberately haven’t created enough projects with Gemini, ChatGPT, or Claude to not be able to switch between them, but with enterprises, things are more complicated and even stronger data moats are in play.
In the enterprise, the data isn’t configured and accessed by a single employee—it is a complex shared resource. Enterprise apps and workflows run on top of it.
Most modern enterprises can be visualized as highly complex, often tangled, dependency graphs between data, people, projects, apps, workflows, etc.
While it’s easy to export data and move it from one SQL database to another, enterprises are so much more complex than that. Products in the enterprise aren’t standalone, and that makes switching much harder.
A specific type of enterprise product we at @2048vc have always been excited about are Systems of Record. We call them data capture machines and continue to feel strongly that certain types of these businesses have very strong moats even in the age of AI.
These pieces of software are essentially highly customized domain-specific databases with workflows, APIs, and plugins built on top. Systems of Record orchestrate complex enterprises by locking down data and delivering apps on top. They are often incredibly difficult to replace, especially in complex categories like EMRs, ERPs, etc.
So let’s make things clear—data moats exist and won’t be going away with AI. Data lock-in is real with both consumers and to a much bigger extent enterprises.
