🎥 Recorded live at the MLOps World | GenAI Summit 2025 — Austin, TX (October 8, 2025) Session Title: Building Effective Agents Speaker: Sushant Mehta, Senior Research Engineer, Google DeepMind Talk Track: Agents in Production Abstract: Large language models can now power remarkably capable software agents, but real-world success requires disciplined engineering rather than flashy frameworks. In this session, Sushant Mehta breaks down what makes agents reliable, scalable, and effective sharing practical patterns and decision frameworks drawn from DeepMind’s applied research. Rather than relying on over-engineered abstractions, he demonstrates how effective agents emerge from simple, composable building blocks and well-timed increases in autonomy. Key discussion points include: - Augmented LLMs (retrieval, tools, memory) as the atomic unit for agent design - Workflow motifs such as prompt chaining, routing, and parallelization—with implementation criteria and tips - Autonomous cycles (plan → act → observe → reflect) for open-ended, self-improving agents What you’ll learn: • A decision framework for evolving from single-prompt systems to multi-step agents • Engineering strategies for maintaining reliability, cost control, and trustworthiness • Design patterns that help teams balance simplicity, autonomy, and performance.