The AI landscape has long been dominated by the “bigger is better” paradigm, but a significant shift is underway. Small language models (SLMs), generally defined as models with fewer than 10B parameters, are emerging as powerful, efficient and practical alternatives to their massive counterparts for a wide range of real-world applications. This talk explores the SLM landscape from the ground up: why they exist, how they are built and trained, what they can and cannot do and where they shine. We draw on recent research, including NVIDIA’s 2025 position paper arguing that SLMs are the future of agentic AI, to make the case that the right model for the job is often a small one. Attendees will leave with a clear understanding of SLM capabilities, architectures, training approaches and the emerging deployment patterns that are reshaping how AI is built and delivered.