Lesson 13.6: Agent Capstone: Choose the Right Agent for the Job

Module 13: Researcher, Analyst, and Microsoft-Built Agents

Lesson 13.6: Agent Capstone: Choose the Right Agent for the Job

Advanced Last verified: 2026-06-02

Mark this lesson complete

Availability note: Researcher, Analyst, installed agents, Agent Store, Agent Builder, knowledge sources, and Copilot Studio paths can vary by license, admin approval, tenant settings, platform, language, and rollout status.

Lesson Promise

Build a complete agent-selection playbook for research, analysis, app work, installed agents, Agent Builder, and Copilot Studio.

Real-World Scenario

A learner must advise a team on which Copilot experience to use for ten real business requests without overpromising agent capabilities.

Core Concept

The capstone combines the module: chat versus agents, Researcher, Analyst, app-specific Copilot, built-in agents, organization-installed agents, Agent Builder, and Copilot Studio.

The best choice is not always the most powerful tool. It is the tool that fits the task, sources, risk, availability, governance, and review needs.

A world-class agent playbook includes selection criteria, rejection criteria, availability checks, data rules, review requirements, and escalation path.

Step-By-Step Workflow

Collect ten representative business requests.
Classify each by task type, source type, repeatability, risk, and output.
Choose the simplest safe Copilot experience for each.
Mark when Researcher or Analyst is appropriate.
Mark when Agent Builder or Copilot Studio should be considered.
Create review and governance notes for each recommendation.

Prompt Lab

Bad Prompt

Tell me which agent to use.

Better Prompt

Create a decision table that maps business requests to Copilot Chat, Researcher, Analyst, app Copilot, installed agents, Agent Builder, or Copilot Studio.

Expert Prompt

Build a complete Copilot agent-selection playbook for these business requests. For each request, include task type, data needed, sensitivity, repeatability, recommended Copilot or agent, why, what not to use, availability checks, admin/governance concerns, source boundaries, output review checklist, and escalation path.

Hands-On Exercise

Create the playbook for ten safe sample business requests.

Deliverable

An agent-selection playbook with decision table, review rules, and escalation path.

Agent Selection Review Checklist

Task type and repeatability defined Sources, permissions, and sensitivity clear Tool or agent is available and approved Assumptions, citations, calculations, and limitations visible Human review owner assigned Escalation path chosen for Agent Builder or Copilot Studio

Common Mistakes

Using an agent because it sounds advanced instead of matching the task to the right tool.
Treating Researcher reports, Analyst reports, or agent answers as final without reviewing sources, assumptions, and limits.
Attaching sensitive data to an agent before checking purpose, permissions, and governance.
Confusing built-in agents, organization-installed agents, Agent Builder agents, and Copilot Studio agents.
Promising agent availability before checking license, admin approval, tenant controls, language, platform, and rollout.

Pro tip: Ask for an agent fit check before using a specialized agent: task fit, source fit, risk, availability, governance, and review needs.

Quiz / Checkpoint

What proves an agent recommendation is strong?

It matches task, sources, risk, availability, governance, and human review needs rather than just choosing the newest tool.

Official Sources To Verify

Previous Course hub Next