Module 13: Researcher, Analyst, and Microsoft-Built Agents
Lesson 13.2: When to Use Chat vs an Agent
Lesson Promise
Choose between ordinary Copilot Chat and an agent based on complexity, repeatability, sources, and output requirements.
Real-World Scenario
A manager wants help drafting a quick update, researching a market, analyzing sales data, and creating a repeatable onboarding helper.
Core Concept
Copilot Chat is often best for flexible conversation, drafting, summarizing, and quick reasoning with selected context.
Agents are better when the task has a specialized purpose, curated knowledge, repeated workflow, deeper research, data analysis, or organization-specific instructions.
A task may start in chat and graduate to an agent when it becomes repeatable, source-specific, or high value enough to standardize.
Step-By-Step Workflow
- Ask whether the task is one-time or repeatable.
- Ask whether the task needs specialized knowledge or tools.
- Ask whether it needs a cited research report, data analysis, or process automation.
- Ask whether the output affects customers, money, policy, or operations.
- Choose the simplest tool that satisfies the task safely.
- Document when the task should be turned into an agent later.
Prompt Lab
Bad Prompt
Should I use an agent?
Better Prompt
This task is recurring and uses our onboarding docs. Should I use Copilot Chat, an installed agent, or build a custom agent?
Expert Prompt
Build a decision tree for choosing Copilot Chat versus an agent. Include one-time tasks, recurring tasks, research tasks, data-analysis tasks, app-specific tasks, knowledge-base tasks, automation tasks, high-risk tasks, and cases where a human should stay fully in control.
Hands-On Exercise
Turn three recurring chat prompts into agent candidates and reject one that should stay as chat.
Deliverable
A chat-vs-agent decision tree with examples and rejection criteria.
Agent Selection Review Checklist
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.
Quiz / Checkpoint
When should a chat workflow become an agent candidate?
When it is repeatable, source-specific, high value, governed, and benefits from consistent instructions or knowledge.
Official Sources To Verify
- Get started with agents in the Microsoft 365 Copilot app
- Get more agents in the Microsoft 365 Copilot app
- Get started with Researcher in Microsoft 365 Copilot
- Get started with Analyst in Microsoft 365 Copilot
- Build your own agent with Microsoft 365 Copilot
- Agent Builder in Microsoft 365 Copilot
- Build agents by using Agent Builder in Microsoft 365 Copilot
- Microsoft 365 Copilot Chat overview
- What Copilot license do I have
Want your AI product explained to a large AI-native audience?
Kingy AI helps AI companies turn complex products into clear, useful YouTube videos that drive awareness, product understanding, demos, clicks, and search visibility.
