AB-730 agents and Cowork explained
Drive business outcomes by using agents is the new AB-730 domain, worth 20–25% of the exam from October 20, 2026. It tests when a prebuilt agent helps, how to choose between a prebuilt and a custom agent, and how to delegate a multi-step task to Microsoft Copilot Cowork and then review what comes back. As a user you apply agents; you are not expected to engineer them.
The outline has two groups: select and apply prebuilt Microsoft 365 Copilot agents, and perform business tasks by using Microsoft Copilot Cowork.
Select and apply prebuilt agents
An agent is a focused assistant configured for a particular kind of job. The skills the outline lists are practical.
| Skill | What to know |
|---|---|
| Identify scenarios | Repeatable work with a clear purpose is a good fit |
| Prebuilt or custom | Start with a prebuilt agent if it covers the need; consider custom if the process is specific to your organisation |
| Apply an agent | Use agents from within various Copilot experiences |
| Filter chats by agent | Find past conversations by the agent used |
The decision pattern is the same in most questions. If the task is common and a prebuilt agent covers it, use it. If the task relies on your company’s own process or knowledge in a way that no prebuilt agent covers, a custom agent is worth considering. The wrong answers tend to be building a custom agent when the prebuilt one fits, or ignoring agents when the task is clearly repeatable.
Perform business tasks with Microsoft Copilot Cowork
Cowork helps carry out multi-step business tasks. Where a chat answers one question at a time, you delegate a piece of work, and Cowork plans and carries out the steps.
| Skill | What to know |
|---|---|
| How it works | Multi-step tasks, using skills to complete parts of a task |
| Prebuilt or custom skills | Prefer a prebuilt skill when it fits; use a custom skill for work that is specific to you |
| Delegating | Hand a piece of work to Cowork with a suitable skill |
| Reviewing | Check its plan, progress and results rather than trusting them blindly |
| Scheduled tasks | Manage tasks or prompts that run on a schedule |
| Security and limits | Know the security considerations and known limitations |
The review step is the theme that links this domain back to fundamentals. Delegating does not move accountability: you remain responsible for the result. Expect questions that ask what to do before accepting what Cowork produced.
Three practice questions
Question 1. A communications team repeats the same kind of task each week, and a prebuilt Microsoft 365 Copilot agent matches it closely. What should the team do first?
- A. Build a custom agent from scratch
- B. Do the task manually
- C. Wait for a future feature
- D. Use the prebuilt agent
Show answer
Answer: D
When a prebuilt agent fits a repeatable task, using it is the most efficient first step. Building a custom agent adds effort without a stated need. Doing the task by hand loses the benefit of an agent. Waiting for a future feature does not help now.
Want more questions like this? Full AB-730 practice tests →
Question 2. A user delegates a multi-step research and drafting task to Copilot Cowork. It finishes and presents a result. What should she do before using it?
- A. Review the plan, steps and results, then check the output
- B. Send it straight to the client
- C. Delete it and start again by hand
- D. Ask Cowork to confirm its own work is correct
Show answer
Answer: A
Reviewing the plan, progress and results is part of using Cowork, and accountability stays with the user. Sending it without review skips that responsibility. Deleting it wastes the work. Asking Cowork to approve its own work is not independent verification.
Want more questions like this? Full AB-730 practice tests →
Question 3. A user wants to find, in her chat history, only the conversations she had with a particular agent. What can she use?
- A. Rename every chat
- B. Filter chats by agent
- C. Schedule a prompt
- D. Delete all other chats
Show answer
Answer: B
Chats can be filtered by agent, which narrows her history to conversations with that agent. Renaming each chat is slow and does not filter. Scheduling a prompt creates future runs. Deleting chats removes the very ones she wants.
Want more questions like this? Full AB-730 practice tests →
How to study this domain
Because the domain is new, older courses skip it. Read Microsoft’s outline bullets for this domain and try each one in the product if you have access. Practise the decision question: prebuilt or custom? The study plan places this domain in week 4.