Kotlin Multiplatform showed us how to share logic. The next frontier is making that shared logic intelligent — and keeping it fast. This advanced session dives deep into how AI agents integrate within modern Kotlin-based Android architectures using Koog, Gemini Nano, and Vertex AI. We’ll explore how to run agents safely and efficiently across Android, iOS, and backend targets — with a strong focus on runtime performance, profiling, and observability. You’ll see how to design and tune agent lifecycles, optimize coroutine-based concurrency for background inference, and measure real-world behavior when running MediaPipe GenAI models locally. We’ll cover how to analyze (profiling) CPU and GPU utilization, memory pressure, and thread contention when agents are reasoning offline — and how to interpret those traces to find bottlenecks in on-device inference. We’ll close with production-ready strategies: structured fallback hierarchies, adaptive inference pipelines, and system-level optimizations that help AI-driven Kotlin apps remain responsive, energy-efficient, and scalable in real-world conditions. By the end, you’ll walk away with a deep understanding of how to embed, measure, and optimize intelligence directly inside your Android codebase — not just integrate it.