In 2022, prompt engineering was the skill everyone talked about. Since then, we've moved from simple chatbots to autonomous, multi-step agents surrounded by a whole ecosystem (RAG, MCP, memory) feeding more and more tokens into our models. Too little, and the model hallucinates; too much, and you get skyrocketing costs, high latency, and the risk of "context rot," where output quality can degrade.
This talk is a practical introduction to context engineering: deciding what your LLM sees, and when. From defining context to concrete optimization techniques for building or using agents (like Claude Code), we'll cover how to keep LLMs on track by giving them only what they need to succeed.
Agenda:
> From prompts to context: what "context" means in the age of agents, RAG, MCP, and memory.
> The balancing act: hallucinations vs. cost, latency, and context rot.
> Practical techniques: controlling context when building or using agents like Claude Code.
