Model Parity Lowers Platform Risk

This points to AI applications becoming less like Facebook games on one dominant platform, and more like software that can switch between several interchangeable model suppliers. The core idea is simple. If OpenAI, Google, and Anthropic are all good enough for many tasks, then an app builder is not fully hostage to any one of them. The real moat shifts up the stack, into workflow, data, distribution, and how well the product turns model output into something useful.

• Platform risk is highest when one upstream provider is uniquely necessary. The interview frames the opposite setup. Google had search index advantages, Anthropic pushed larger context windows, and OpenAI led on reasoning, which meant capability was spreading across multiple vendors rather than concentrating in one.
  1 sacra 2 openai
• This matters for products built on top of models because they can design for swapability. That pattern shows up elsewhere in AI infrastructure. Ecosia built its Exa integration to stay flexible and avoid deep lock in, because the market was moving fast and providers were becoming easier to substitute.
  6 sacra
• For Reddit specifically, lower model concentration makes its content asset more valuable, not less. If several labs need fresh human discussion data, Reddit can license to more than one buyer instead of depending on a single winner. That fits the broader view of Reddit as a source of net new, niche, high intent text for AI training and products.
  3 sacra 4 sacra 5 sacra

The next phase pushes model providers toward utility pricing and commodity economics. As parity improves and open source gets good enough for more tasks, the durable winners at the application layer will be the companies that own user behavior, proprietary data, and repeat workflows. In that world, models are still crucial, but they look more like cloud infrastructure than the whole business.