Notes: Get Better Agent Results With Prompt Engineering

Section 6 develops a complete request through Daniel's project-management comparison, then applies the same method to Arun's customer-group research. The examples connect the job, prompt, visible activity and deciding evidence. Prepared requests support optional personal practice.

Lecture	Practical development
How Can You Get Useful Work From AI? (A Worked Decision)	The earlier recorded project-management shortlist needs Daniel's group-cost and private-note checks before it can support his three-person decision.
Which Job Should You Give an AI Agent First? (Use Cases)	Daniel's project-management comparison shows why an investigation is a useful first use case.
Which Tasks Make a Poor First Agent Exercise? (Task Selection)	Four first-run difficulties explain why Daniel narrows his task, supplies requirements and keeps purchasing outside the research.
What Does an AI Agent Guess When Your Prompt Is Vague?	Daniel's vague request reveals facts the application must ask about, assume or omit. The Claude Code result remains a proposed capture.
How Do You Write a Prompt for an AI Agent? (Requirements)	Purpose, scope, criteria, tradeoff, output format, sources and a stopping boundary become one runnable message.
Which Changes Improve an AI Prompt? (Few-Shot and Chain-of-Thought)	Examples teach a missing response pattern before zero-shot, one-shot and few-shot are named. Checkable evidence replaces a demand for private reasoning.
How Can You Tell a Good AI Answer From a Bad One? (Checking Results)	Fictional Tools A, B and C let the learner rehearse eliminating an unsuitable option, comparing total costs and preserving the export tradeoff before applying the order to a later result.
Can an AI Agent Help With a Business Decision? (Claude Code Case Study)	The exact RegulatorySense customer-group request demonstrates how Claude Code can support a human business decision. The run and answer remain proposed captures.
What Is an AI Agent Doing While You Wait? (Latency and Activity)	Latency and visible activity help the learner decide whether to wait, answer a permission request, redirect drift or preserve an error.
When Should You Interrupt an Agent?	Judge an unexpected action by the requirement it helps answer. Redirect a misunderstanding, interrupt repeated failure, and preserve useful work before restarting.
How Does an Agent Plan Its Work? (ReAct and ReWOO)	ReAct uses returned information to help choose the next action. ReWOO separates planning requests from collecting results. These names do not identify Claude Code’s private method.
Why Did Your AI Agent Stop? (Errors and Usage Limits)	The exact error distinguishes an unclear request, unavailable source, failed approach or stated usage limit and guides one relevant correction.
Can You Explain an AI Agent Run? (Prompt, Evidence and Judgment)	The worked review connects a request and result with deciding facts, sources, unresolved points and a useful correction. A personal record is optional.

A useful review explains which evidence supports a recommendation and what remains unresolved. If you run a task yourself, its existing conversation can keep the request, answer and correction together.