In this video, I’m talking about an AI bubble that I don’t think enough people are pricing in: the “wrapper” problem. A huge chunk of funded AI startups are essentially built on top of OpenAI or Anthropic. That’s not automatically bad—wrappers can still deliver real value—but it creates a fragile business model when your core capability is rented from someone else. The big issue is valuations. A lot of these companies are being valued as if LLM costs will stay low forever (or keep dropping). But the model providers themselves are burning money. If prices go up, margins get crushed overnight. And if one of these providers hits a wall—policy change, pricing change, capacity limits, or a collapse—then entire categories of products get exposed because they don’t control the underlying engine. I’m not claiming I know exactly how this plays out. This is speculation—but it’s worth having the conversation. If you’re building, investing, or even just choosing tools for your workflow, you should be asking: what happens to this product when tokens get expensive? That question alone changes how I think about “AI productivity” tools and which ones are actually durable.