Talk 01
Foundations of AI-Powered App Architecture
Building AI-powered applications requires more than connecting a language model to an interface. This session introduces the foundational concepts behind modern AI apps, including prompt structure, context management, and system orchestration.
A central focus of the talk is agent-based architectures: how agents are constructed from models, tools, memory, and control loops, and which patterns are commonly used in practice. It also covers evaluation strategies for probabilistic systems, and the common challenges with agent systems, such as loss of control, debugging difficulty, and cost explosion, and how to design around them.
What the audience takes away
- A clear mental model for building AI-powered applications.
- How agents are constructed from models, tools, memory, and control loops.
- Ways to measure task success, faithfulness, cost, and reliability.
- Practical guidance for making agent systems evaluable and observable.