The notion of building a business as a 'thin wrapper' around a foundational model like GPT is flawed. Truly defensible AI products, like Cursor, build numerous specific, fine-tuned models to deeply understand a user's domain. This creates a data and performance moat that a generic model cannot easily replicate, much like Salesforce was more than just a 'thin wrapper' on a database.

For long-term defensibility, AI companies must control the entire stack: the model, the middleware, and the end-user work product. While some can start with the model layer, others can successfully start with the user interface and vertically integrate downwards over time to build a durable business.

Similar to how blockchain protocols like Bitcoin and Ethereum accrued more value than the apps built on them, AI foundation models are getting 'fatter.' They are absorbing more capabilities, allowing users to perform complex tasks in a single step within the base model, reducing the need for specialized application-layer companies.

Counter to fears that foundation models will obsolete all apps, AI startups can build defensible businesses by embedding AI into unique workflows, owning the customer relationship, and creating network effects. This mirrors how top App Store apps succeeded despite Apple's platform dominance.

Gurley notes that major AI model providers like OpenAI and Anthropic are shifting from solely selling API access to building their own applications. This move up the stack signals a fear that being a pure model provider is not a defensible moat and could lead to commoditization.

The enduring moat in the AI stack lies in what is hardest to replicate. Since building foundation models is significantly more difficult than building applications on top of them, the model layer is inherently more defensible and will naturally capture more value over time.

In the AI era, defensibility comes from building a complex system of record, not just a thin wrapper on an LLM. Companies with a 'thick application layer' that offers standalone value are unattractive for model providers to replicate, whereas thin wrappers risk being absorbed by the platform they are built on.

Startups building on top of AI models, like coding assistant Cursor, are extremely vulnerable. As foundation model companies like Anthropic improve their own native capabilities (e.g., Claude Code), they can quickly capture the market and render specialized tools obsolete.

The battleground for AI startups is constantly shrinking like the map in Fortnite. Foundation models like Anthropic's Claude are aggressively absorbing features, turning what was a standalone product into a native capability overnight. This creates extreme existential risk for application-layer companies.