Session: The Context Economy: Getting more bang for your token
Building effective agents requires fundamental changes to how engineers think about their work. LLMs that sit at the heart of agentic systems are stochastic by nature, and operate in a world that requires optimal context to produce optimal results. While models offer bigger and bigger context windows, the cost and performance during inference is tied to the consistency of context delivery and the stability of that context over time. As tool calling and the Model Context Protocol (MCP) ecosystem play a larger role in agentic development, engineers need to invest in not only delivering the most semantically relevant information to the task at hand, but maintaining those systems in a constantly changing and evolving world.
During this session, we will dive into the operational best practices at the intersection of agentic workflows and tool calling. We will look at practical approaches to optimize tools as a natural part of agent development, and discuss approaches to optimize the most critical resource in agentic development: context real estate.
