Rethinking Scale: Deployment Trade-offs of Small Language Models under Agent Paradigms | ScienceToStartup