A few months ago, the CEO of Cursor, one of the most successful AI coding tools on the planet, built on top of models from the very labs it now competes with, said something remarkable out loud. He warned that Anthropic's own coding agent, Claude Code, could "bring Cursor's growth to a crashing halt."
Sit with that for a second. A company that raised hundreds of millions of dollars, built a product millions of developers genuinely love and did everything right, got beaten, not by a smarter competitor but by its own supplier deciding to compete with it directly, using the very model Cursor was renting.
That's not a cautionary tale about Cursor. Cursor is good at what it does. It's a cautionary tale about where you choose to put your foundations.
The pitch everyone's making right now
Right now, the loudest advice in tech is; just build on the model. Point your product straight at a foundation model's API, let it do the thinking, ship fast, iterate faster. And for a huge amount of what gets built today; prototypes, internal tools, weekend projects, genuinely disposable software, that advice is completely right. Speed wins.
But look closely enough at the pattern, not just Cursor but the growing pile of stories just like it and an uncomfortable truth comes into focus: there’s a real difference between building with AI and building on AI and most of the advice flying around right now treats them as the same thing.
When you build on AI, when your product's core logic, your customer data, your competitive advantage all live inside someone else's model, called through someone else's API, orchestrated on someone else's platform - you have made a decision you probably didn't mean to make. You decided that your company's future is a feature of someone else's roadmap.
And the labs building these models are not passive infrastructure providers, they have interests of their own. They're some of the most ambitious, best funded, fastest moving companies in the history of software and they are increasingly building the exact same layer you're building on top of.
We’ve seen this playbook before, just with a different landlord. Amazon built a marketplace for other sellers, then became a seller itself, competing on the very platform those businesses had helped make successful. The more you build on someone else’s platform, the more control you hand to someone who has no obligation to protect your position.
What you actually lose
It's not only the risk of a competitor showing up wearing your supplier's badge, though that's real enough on its own. It's three quieter things.
You lose visibility. When your product runs in someone else’s environment, they can see how it performs in the real world and use those insights to improve their own models and competing tools. You, meanwhile, see your inputs and outputs and not much else.
You lose leverage. Pricing changes, usage policies change, models get deprecated, "free" tiers quietly stop being free. None of that is hypothetical anymore, it's the normal rhythm of this industry. If your entire product sits on one vendor's foundation, you don't get a vote.
And you lose the thing that actually makes a business valuable over time; a moat that's yours. If anyone with API access to the same model could rebuild the clever part of your product in a weekend, the clever part was never really your product. Your product was always meant to be everything around the model; the data, the workflows, the judgment, the relationships, the specific problem you understand better than anyone else.

Own the core. Rent the clever bit.
Here's the shift in thinking we keep coming back to, with our own ventures and with the clients who bring us in to build. Treat the model as a component, not a foundation. A brilliant, fast moving, occasionally game changing component, but a component. Something you can swap out, upgrade, or route around, the same way you would treat any vendor whose roadmap you don't control.
A useful test: if your AI provider 10x its pricing tomorrow or deprecated the model you rely on, or launched a product that does what you do, would your business survive the week? If the honest answer is "no, we'd have to rebuild from scratch," you haven't built a company. You've built a very elegant wrapper around someone else's decision making.
The companies that will still be standing in five years are quietly doing the less glamorous work right now: owning their data and their context, keeping their core architecture model agnostic and building the parts of the business the domain expertise, the customer trust, the specific workflow nobody else has bothered to understand as deeply… that no model update can take away from them overnight.
AI is the best tool we've ever had for building fast. It's a terrible place to build your foundation. Build the house on your own ground and let AI be the best contractor you've ever hired, not the landlord.
“AI is the best tool we've ever had for building fast. It's a terrible place to build your foundation.”




