Building and Optimizing AI Agent Workflows

This long course equips you with practical knowledge and hands-on skills required to design, architect, and optimize autonomous AI agents that solve multi-step tasks reliably, efficiently, and responsibly. You will study reward-design and reinforcement-learning foundations to translate business objectives into robust reward signals, while learning to evaluate ethical, legal, and societal impacts of agent decision policies. The course covers competing reasoning-loop architectures (e.g., ReAct and Reflexion), modular agent component design with clear APIs, and search and planning strategies (A*, beam search, and heuristic augmentation). You will also practice feature engineering and model-interpretability methods to expose spurious correlations and produce explainable agent behaviors. Finally, the course guides you to make strategic modeling choices—such as fine-tuning large models versus training smaller task-specific models—and to package reproducible, reusable ML pipelines for agent subsystems. Throughout the course, practical labs and engineering-focused examples emphasize production-readiness, modularity, and trustworthiness.