Intro to Agentic Design Patterns

In this lab you build two common agent architectures in plain Python, without a framework. The first notebook covers orchestrator-worker: an orchestrator splits a review task across specialized workers and combines their results. The second covers ReAct: a single agent reasons about a question, calls a tool, reads the result, and repeats until it has an answer. In both notebooks you run the system on tasks it handles well, then on one it gets wrong, and check the results with code rather than by reading the model's output.

Goal: Teach the orchestrator-worker and ReAct patterns by building each in plain Python and driving it into the failure it is structurally blind to.

Learning Objectives - by the end of the lab, you will be able to:

• Implement an orchestrator that routes documents to scoped workers and combines their results
• Explain why a problem spanning two documents is invisible to workers that each see one
• Implement a ReAct loop with schema-validated steps and read its trajectory
• Explain why an unanswerable question drives the loop to its iteration cap
• Add a structural stall guard that stops the loop without relying on the model to notice

• Kong AI Gateway for LLM Access
• Ubuntu Jumpbox running Jupyter