Omar Khattab challenges the prevailing obsession with larger language models, arguing that true progress lies in building programmable, reliable AI systems. His framework, DSPy, introduces formal abstractions like 'signatures'—declarative contracts that separate user intent from model execution. This shift enables modular, composable AI applications that remain stable even as underlying models evolve. Rather than relying on brittle natural language prompts or imperative code, DSPy blends typed structures with optimization techniques, allowing developers to specify *what* should be achieved without dictating *how*. As frontier labs move beyond pure scaling, investments in retrieval, tool use, and agent systems highlight a broader industry pivot toward structured design. The future of AI, Khattab asserts, depends not on emergent magic but on engineering principles that make systems interpretable, maintainable, and aligned with complex human goals.