Phase 2 · Microsoft Copilot · Level 3 · Power User
Microsoft Copilot: Power User Quiz · 10 min
Why it matters A quick check of your judgement before you move on. There's no penalty for a low score, and you can retake it, or take it first to test out of a level you already know.
These scenarios test the judgement of a Copilot power user: building agents, knowing where Copilot Studio and agentic tasks fit, understanding what Copilot can and can't see, and handing off to Power Automate. They're situations, not trivia: pick the best option and read the explanation after each.
Passing this checkpoint: work through the level first, then score 70% or more here.
Skipping this level: already confident? Take this cold and score 80% or more to test out and jump ahead. Below that, nothing is lost, you just study the level as normal.
1. Your team keeps re-pasting the same policy document and the same instructions into Copilot to answer staff questions about remote working. What's the power-user move? Write the perfect prompt once and email it round for everyone to copy. Build a declarative agent in Agent Builder, grounded on the policy document, with instructions to answer only from it, so anyone can open it by name instead of rebuilding the prompt each time. Ask Copilot to memorise the policy so no one has to paste it again. Switch to Copilot Studio, because any reusable assistant needs it. 2. You build a declarative agent in Agent Builder and give it a policy document as its only knowledge source. To trust it before sharing, what's the most important test? Check it answers questions the policy covers; if those are right, it's ready. Test both that it answers covered questions accurately AND that it refuses to invent an answer when you ask something the policy doesn't cover. Ask the agent whether it's reliable and share it if it says yes. Confirm it responds quickly, since speed is the main quality signal. 3. You want to ground an agent on a large, living body of reference material a team updates constantly, so answers stay current. Better grounding choice? Upload a snapshot of the individual files today and re-upload them whenever they change. Point the agent at the whole SharePoint site, so it searches the live content and stays current as the team updates it. Paste all the documents into the instructions box. Use no sources and rely on Copilot's general knowledge of the topic. 4. A colleague says 'I need my agent to answer a question AND then log a ticket in our system'. Which tool does that push them towards, and why? Stay in Agent Builder, since it can take actions like creating records. Copilot Studio, because the 'and then do something' part needs actions, logic and connectors that go beyond Agent Builder's answer-from-sources scope. An agentic task, because anything with two steps is agentic. Power Automate only, because Copilot can't be involved in creating records at all. 5. You give Copilot a long agentic task (Cowork-style) that runs for 40 minutes across several files. It shows visible progress as it works. What's the visible progress actually for? It's a loading animation, nothing to act on, just wait for the final output. It's your chance to steer and catch a wrong turn early, because a mistake at minute two gets built on by every step after it. It proves the task is accurate, since you can see it working. It means the task can safely run unattended once it's started. 6. A manager worries Copilot will 'leak confidential files across the company'. Based on this level, what's the accurate reassurance, and the real risk to name? Copilot can indeed access anything in the tenant, so the worry is justified; restrict it heavily. Copilot only ever shows a person content they already had permission to open; the real risk is that it efficiently surfaces content that was already overshared, so the fix is a permissions review. Copilot invents a random subset of files to show, so results are unpredictable. There's no risk at all, so no permission review is ever needed before a rollout. 7. Before rolling your agent out to a whole department, you want to make sure sensitive documents can't be used in its answers even where permissions are loose. Which governance lever is that? There's no way to stop Copilot using a document someone can technically see. A Microsoft Purview DLP policy (and sensitivity labels) can tell Copilot not to use labelled content when generating answers; the item may still appear as a citation, but its content isn't read into the response. You delete the sensitive documents entirely so no one can use them. You ask each user politely not to query those documents. 8. Every time a form is submitted, you personally ask Copilot to save the attachment and post a note in the team channel. It's identical each time. What should you do instead? Keep prompting Copilot each time; it's already handling it fine. Build a Power Automate flow: the form submission is the trigger, and saving the file and posting the note are the actions, so it runs unattended, every time, without you. Write a longer Copilot prompt so it does more of the steps at once. Turn it into a declarative agent, since agents handle repeated tasks. 9. Power Automate's Copilot drafts a flow from your plain-English description in seconds, setting up the trigger and connections. What's the right level of trust for the first run? Full trust: Copilot built it, and a flow that runs unattended is reliable by design. Check the drafted trigger, action mappings and connections, then test it on a sample and watch the first real run, because a wrong mapping repeats silently every time until someone notices. No trust: Copilot-built flows never work and should be rebuilt by hand. Trust it only if the description was longer than three sentences. 10. You've built an impressive grounded agent and it wowed you in testing. Writing the one-page rollout note for your manager, what's the power-user discipline? Present it as fully reliable, since you built it and it impressed you, so talk up its strengths. Let the evidence set the claim: report both test results honestly (including that it needed two rounds to stop guessing), name who can see the sources, and state where a human still checks. Fill the note with the build steps and exact prompt wording so the manager can rebuild it. Leave out governance, since that's IT's job, not something for a rollout note. Answered 0 of 10.
Answer every question to see your score