Context Engineering

What is context engineering?

How do you handle conversations that exceed the context window?

What is the 'lost in the middle' problem?

What is prompt caching and how does it save cost?

How do you design memory for a long-running AI system?