Georgia State University — J. Mack Robinson College of Business PATH — Pathways for AI Training & Hiring CIS 4394 Agentic AI  ·  Fall 2026  ·  Dr. Xinyu Fu
Week 3 · Foundations

Prompting & Context Engineering.

How to make a probabilistic model behave like a reliable business tool. Last week you drove a coding agent and saw its harness. This week we zoom into one layer (the instruction layer) because it is the highest-leverage, lowest-cost skill in the entire stack.

This week:
  • 📝 Quiz 1 (Weeks 1–3 foundations, 5 multiple choice, closed-book) at the end of class; readings and sample questions are on the Week 2 Summary page.
  • 👥 GA1 is due this week; check iCollege for your section’s due date, and join a group if you haven’t.
  • 📡 Agent Radar opens class.
The hook

One prompt change flips reliability

Same model. Same 20 messy vendor emails. The only thing that changes is the prompt. (Illustrative teaching example; you’ll reproduce this effect yourself in the hands-on.)

Before
What’s the total?
Usable outputs: sometimes a number, sometimes a sentence, sometimes the wrong currency…
~60% usable · format varies every run
After
SYSTEM: You are an invoice-processing assistant. Return ONLY JSON: {"total_usd": number|null} If no total is present, use null. EXAMPLE input: "…amount due USD 1,284.50…" output: {"total_usd": 1284.50}
Same model: now every answer is a clean, parseable record.
~100% usable · identical shape every run
The gap between ~70% and ~98% reliability is the gap between a tool people trust and a tool people turn off. Prompting is free and instant; it’s where an MIS professional adds value without an ML PhD.
The cost lever

Cheapest fix first

When an LLM feature underperforms, there is an ordered menu of fixes, and the price rises steeply as you go down it.

FixCostSpeedWhen
1 · Edit the promptFreeSecondsAlways start here
2 · Add few-shot examplesFree (a few tokens)MinutesFormat or edge-case misses
3 · Better context / retrievalEngineering timeDaysModel lacks the facts (Week 7: RAG)
4 · Fine-tuneData + compute + upkeepWeeksConsistent style at scale, hundreds+ examples
5 · Wait for a new model—QuartersNot a strategy
See it move · the context window as a budget

What falls out when you stuff the window

Everything the model knows about your task has to fit in one window. Press Play and watch the pieces arrive: system prompt, examples, retrieved documents, chat history, the new question. When the window is full, the oldest turns are evicted, and the answer quietly changes.

Try this: in “Overflow”, note which piece carried the fact the answer needed. Then in “Curated”, say in one sentence what the summary step traded away to keep it.

This week

Work through the four parts

Objectives

By the end of Week 3 you can…

← BackCourse home