Loop Engineering for FDEs

An agent is an LLM running tools in a loop. Run it long enough and you often get nonsense.

Why? You hand a coding agent a task and walk away. Forty turns later, it is debugging its own confusion.

The fix isn’t a better model. It’s a better harness: the code around the agent that decides what it sees, checks its work, and stops it.

This workshop teaches loop engineering, a hands-on slice of harness engineering, so you can actually walk away.

Part I: Loop Engineering with Deep Agents

As in The AI FDE Certification, we start with a real business use case and build it with a leading framework, LangChain’s Deep Agents.

Part II: Loop Engineering from Scratch

Then we rebuild it in plain Python using Ralph Loops and Spec-Driven Development. FDEs need both: in a large enterprise, writing the code yourself is often faster than getting a new library approved.

Building it both ways shows what a framework gives you, and what it never will: the stop condition is yours to write.

Half the day is hands-on. You’ll work through both notebooks in small groups, then write a spec and checker for your own work with Dr. Greg and Mr. Chris in the room.