Your email AI agent works perfectly in demos but fails in production. It confuses client contexts, suggests outdated meeting times, and drowns in month-old thread history. You manually fix prompts after each failure, but the problems keep evolving.

 

The real issue? Static context management can't handle email's dynamic complexity.

 

In this 50-minute hands-on session, we'll build an email agent that combines context engineering with DSPy's self-optimization. You'll implement the four-move context framework: Write, Select, Compress & Isolate, but supercharged with optimizers that learn from every correction you make.

 

Instead of hardcoding rules for "include last 5 emails" or "summarize threads over 1000 tokens," your agent will learn what context actually matters for YOUR email patterns. When you correct a response, DSPy optimizers automatically adjust how the agent selects, compresses, and isolates context: with minimal prompt engineering.