The inconsistency and 'laziness' of base LLMs is a major hurdle. The best application-layer companies differentiate themselves not by just wrapping a model, but by building a complex harness that ensures the right amount of intelligence is reliably applied to a specific user task, creating a defensible product.

Don't just sprinkle AI features onto your existing product ('AI at the edge'). Transformative companies rethink workflows and shrink their old codebase, making the LLM a core part of the solution. This is about re-architecting the solution from the ground up, not just enhancing it.

The most defensible AI companies don't just have superior models; they embed themselves deeply into customer workflows. The primary barrier to adoption is change management, so overcoming that hurdle creates a durable competitive advantage that is difficult to displace.

As AI commoditizes user interfaces, enduring value will reside in the backend systems that are the authoritative source of data (e.g., payroll, financial records). These 'systems of record' are sticky due to regulation, business process integration, and high switching costs.

Strong AI products require a tight feedback loop where the product and model are deeply integrated. Thin wrappers around third-party models create weak, short-lived features that will be subsumed by the platform. A durable AI business treats the model *as* the product itself.

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.

As AI application layers become easier to clone, the sustainable competitive advantage is moving down the tech stack. Companies with unique, last-mile user interaction data can build proprietary models that are cheaper and better, creating a data flywheel and a moat that is difficult for competitors to replicate.

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.

An AI app that is merely a wrapper around a foundation model is at high risk of being absorbed by the model provider. True defensibility comes from integrating AI with proprietary data and workflows to become an indispensable enterprise system of record, like an HR or CRM system.