A new model drops. It's better at benchmarks. Every AI product company now has to integrate it (if they're a wrapper) or match it (if they're building a model) or explain to their investors why they didn't. Integration // parity takes engineering time, time that doesn't go toward the actual product differentiation that would create long-term value. But the company that doesn't integrate looks like it's falling behind, and looking like you're falling behind is fatal when your entire valuation rests on the assumption that you're on the frontier.

This paragraph very nicely highlights the poison of a financialized economy. Where the ability to attract capital is more important than having an actually useful product.