Lesson 14.2: Good Agent Ideas vs Bad Agent Ideas

Module 14: Agent Builder

Lesson 14.2: Good Agent Ideas vs Bad Agent Ideas

Beginner Last verified: 2026-06-02

Mark this lesson complete

Availability and governance note: Agent Builder availability, builder permissions, sharing, publish options, and supported knowledge sources can vary by tenant, license, admin controls, and rollout.

Lesson Promise

Separate strong agent candidates from risky, vague, or unnecessary ones.

Real-World Scenario

A leadership team proposes ten agents, including a policy helper, a sales-call coach, a legal advisor, and a company-wide truth machine.

Core Concept

Good agent ideas are narrow, repeatable, source-backed, low to moderate risk, and easy to test against real user questions.

Bad agent ideas are vague, high authority, unsupported by clean knowledge, or likely to make regulated decisions without a human.

An idea that is too broad for Agent Builder may still become a Copilot Studio project if it needs channels, actions, authentication, analytics, or lifecycle controls.

Step-By-Step Workflow

Score each idea for repeatability, source quality, sensitivity, risk, and testability.
Reject ideas that ask the agent to make final legal, medical, HR, finance, or compliance decisions.
Narrow broad ideas to one audience and one workflow.
Choose Agent Builder for simple knowledge help and Copilot Studio for richer process agents.
Define pass/fail tests before building.
Document who owns accuracy and updates.

Prompt Lab

Bad Prompt

Build an agent for everything employees need.

Better Prompt

Build an agent that answers approved PTO policy questions and routes unusual cases to HR.

Expert Prompt

Evaluate these agent ideas for Agent Builder readiness. Score each for business value, repeatability, source readiness, sensitivity, risk, testability, owner, and whether it should be chat, Agent Builder, Copilot Studio, or rejected.

Hands-On Exercise

Rank five proposed agents and explain which two you would build first.

Deliverable

An agent idea scorecard with build, defer, escalate, or reject decisions.

Agent Builder Review Checklist

The agent has one clear job and a named audience. Instructions define tone, source boundaries, refusal behavior, and escalation path. Knowledge sources are approved, current, and permission-safe. Conversation starters represent common real work, not demo-only prompts. Testing includes expected answers, bad inputs, missing information, and sensitive requests. A human owner is assigned for content updates and user feedback.

Common Mistakes

Building a broad general helper instead of a focused role-based agent.
Adding messy or outdated files and expecting the agent to infer what matters.
Skipping refusal instructions for policy, legal, financial, HR, or customer-impacting questions.
Testing only the happy path and missing vague, hostile, or sensitive prompts.
Sharing an agent before ownership, feedback, and update routines are clear.

Pro tip: Write agent instructions like operating policy: purpose, audience, allowed sources, preferred structure, quality rules, and what to do when the answer is uncertain.

Quiz / Checkpoint

What is the biggest warning sign in an agent idea?

A broad or high-risk agent with unclear sources, unclear owner, and authority to make final decisions.

Official Sources To Verify

Previous Course hub Next