2.3. Chat, Agents, and Copilot Cowork
💡 First Principle: Think about how you delegate to people. You ask a colleague a quick question. You hand a specialist a deep piece of analysis. You give an assistant a job to do: send the invites, book the room, file the report. Copilot now has an option for each level. Copilot Chat answers and drafts, agents specialize, and Microsoft Copilot Cowork acts. Pick by asking what kind of delegation the task needs.
Choosing the wrong option costs time, quality or safety. Use chat for a 30-page cited competitive analysis and you get a shallow answer. Use Researcher when you needed the follow-up emails sent and the meeting booked, and you get a report but nothing done. Use Cowork for a one-line rewrite and you start a multi-step session (with usage-based billing) for something chat does in seconds. The exam keeps testing this choice because all three live in the same app and look alike at first glance.
| Copilot Chat | An agent (for example Researcher) | Copilot Cowork | |
|---|---|---|---|
| What it is | Always-on AI for drafting, summarizing and answering | A scoped AI experience built for one kind of job | An agentic system that completes multi-step work across Microsoft 365 |
| Shape of the work | Single task, single step, single session | Deep work within its specialty | End to end, across apps |
| Typical duration | Seconds to minutes | Minutes (Researcher can take longer) | Minutes to hours |
| Takes actions in your apps? | No, you act on the output | Generally no, it produces an output | Yes: sends, schedules, posts and creates files, with your approval |
| Typical scenarios | Daily catch-up, project status, Q&A, drafting | Cited research report, data analysis | Inbox and calendar clean-up, project launches, meeting prep |
Microsoft's own summary is a good one to memorize: "Copilot Chat helps you think through your work… Cowork helps you get work done." Microsoft's admin documentation also stresses that Cowork is an agentic system, which is different from an agent.
Agents come in two kinds. Prebuilt agents are made by Microsoft (Researcher, Analyst, the coaches and others) and work the same way for everyone. Custom agents are built by someone in your organization for a specific purpose, such as a team FAQ agent over a project site. Phase 6 covers how to choose between them. For now, the key idea is that an agent is a different kind of experience, not just a smarter chat.
⚠️ Common Misconception: "Researcher is the most powerful option, so it can handle the whole workflow." Researcher researches. It produces a cited report, but it doesn't send the summary email or book the follow-up meeting. When a scenario asks for work to be carried out across apps, the answer is Cowork. When it asks for a deep answer, it's an agent like Researcher. When it asks for a quick draft, it's chat.
