2.3. Chat Experiences vs. Agent Experiences
💡 First Principle: A Copilot chat experience is a general-purpose conversation with an AI assistant that draws on its training and whatever context you provide. A Copilot agent experience is a pre-configured, scoped AI assistant with specific instructions, knowledge sources, and capabilities — it behaves differently because it was designed to behave differently. Choosing the wrong experience costs you quality and safety. Ask a general Chat for your company's current leave policy and you get generic training-based text that may be outdated or wrong for your organization. Ask a purpose-built HR agent and you get an answer grounded in your actual policies — and bounded from answering things outside its scope. Without understanding this distinction, users default to Chat for everything and miss both the precision and the guardrails that agents deliver.
The distinction between chat and agents is one of the most heavily tested concepts on AB-730. Many candidates conflate them because both use the same Copilot interface — but they serve fundamentally different purposes.
When to use chat vs. an agent:
| Scenario | Best Choice | Why |
|---|---|---|
| Draft a quick email response | Chat | General task, no specialized knowledge needed |
| Answer questions about your company's HR policies | Agent (configured with HR policy docs) | Needs specific, scoped knowledge |
| Summarize an article you found online | Chat | General task |
| Help employees navigate IT support procedures | Agent (IT support agent) | Consistent, scoped responses from IT documentation |
| Brainstorm marketing ideas | Chat | Open-ended creative task |
| Guide new employees through onboarding steps | Agent (onboarding agent) | Structured, repeatable workflow |
What makes an agent different from chat:
- Instructions: An agent has a system prompt that shapes how it responds (e.g., "Always respond in formal English" or "Only answer questions about our product catalog")
- Knowledge: An agent is configured with specific knowledge sources (SharePoint sites, files, websites) it draws from
- Capabilities: An agent can have specific capabilities enabled (search, actions, connectors)
- Suggested prompts: An agent can surface pre-written prompts to guide users toward common tasks
Understanding when to create your own agent (rather than using the Agent Store): Build your own when you have a specialized, recurring use case that no existing agent addresses — for example, an agent that answers questions specifically about your organization's proprietary processes using your internal documentation.
⚠️ Exam Trap: An agent is not just a smarter version of chat. It is a different kind of experience. An agent you create does not automatically have access to all Copilot knowledge — it only uses the knowledge sources you explicitly configure. If you want it to know about your company's benefits policy, you must add that document to the agent's knowledge.
Reflection Question: A team lead wants Copilot to always answer employee questions about project timelines using only the official project tracker spreadsheet. Should they use Copilot Chat or create an agent? Why?
