Watch: SEO is fine. The future of AI is tied to search.

LLMs function best as reasoning engines rather than static databases, relying on search indexes and massive context rather than model weights to access facts.

Transcript

Large language models are not all-knowing databases. Using an AI's weights to store facts is slow, expensive, and incredibly wasteful. Those weights are far better used to compress common sense and general intelligence. The real power of these models is reasoning, not memorization. Even Sam Altman envisions a future with a tiny, ultra-fast reasoning engine rather than a massive, static database. In this ideal state, a model never needs to retrain. Instead, it has a massive context window—perhaps a trillion tokens—where you can stream your entire life, your business data, and any tool you need, letting the model reason across all of it in real time. One company already has the perfect setup for this: Google. Their models don't need to know everything on their own because organic search indexes supply the real-time memory. The language model simply acts as a presentation layer, while search and integrations like Chrome supply the live context. Here at DEJAN we see this reality shaping the future of search and AI. It is not about building a bigger database. It is about perfecting the reasoning engine.