Context Engineering

The instructions, memory, tools, and constraints we provide to agents ultimately define their behavior, reliability, and usefulness. However, in many teams, context is still treated as an ad-hoc layer: prompts are scattered, assumptions are implicit, and behaviors are difficult to reproduce or scale.

This talk introduces Context Engineering as a core discipline in agentic system design. It reframes context as a structured, testable, and reusable artifact, one that can be designed with the same rigor as software systems.

This session explores how to design context that is robust across tasks, portable across models, and maintainable across teams. In this session, attendees will learn how to:

• Structure agent context intentionally by defining specifications, rules, and capabilities that guide behavior beyond single prompts.
• Design reusable context systems that can be shared across tasks, agents, and teams without breaking under variation.
• Evaluate and iterate on context as a first-class artifact using testing, tracing, and feedback loops.
• Apply distribution patterns for context including versioning and cross-team sharing.
• uild for multi-model environments by making context portable, adaptable, and resilient to model differences.

‍