Generative Agents are everywhere in the news, but moving past the hype and into reliable, real-world applications is the hard part. Building an agent that reliably solves non-trivial problems is a significant engineering challenge.
This talk dives into the practical engineering side of building effective AI agents, focusing on design patterns. We will break down the essential components and compositional strategies that allow an agent to move beyond a simple LLM wrapper and into a robust, goal-directed system.
You will learn:
-
Essential Agent Design Patterns (the "compose patterns") for constructing complex reasoning, memory, and tool-use.
-
Patterns in Practice: Three agents as case studies demonstrating how these architectural patterns overcome real-world hurdles.
-
Development Best Practices around iterative development, areas of focus, and agent evaluation.
Jeff Fischer is a Senior Director of Product Management. He focuses on building Agentic AI application platforms. He's an engineer at heart who's spent years diving deep into software systems, programming languages, and data science. He holds a PhD from UCLA and has led engineering and product teams at companies large and small.
- Slides for talk at BayPiggies Nov 20, 2025
- A longer version of the talk
- A "vibe-coded" agent framework
- Open source agents
- Some relevant papers you might want to read for more background