Your staff don’t need AI agents explained to them yet. They need to know what they’re allowed to use Copilot for, and they need one piece of their own work where they can check the result with their own eyes.
Workday surveyed nearly 6,000 workers around the world in September 2026, and only 38% of employees said they understood the term AI agent well enough to explain it confidently. Senior leaders were 2.5 times more likely than individual contributors to say they understood it.
How wide is the gap between leaders and staff?
The gap runs through every measure below. These are the Workday AI@Work figures for September, and they’re global, so I’m not claiming anything about Australia specifically.
| Senior leaders | Individual contributors | |
|---|---|---|
| Use AI in their work | 95% | 62% |
| Heavy AI users | 44% | 9% |
| Have used agentic AI to complete a task | 88% | 54% |
| Trust their employer’s approach to AI | 66% | 32% |
If you’re a senior leader, you’re probably in the left column. You’ve handed whole tasks to AI and seen them come back done. A lot of the people you’re talking to are still working out what Copilot can do in the document in front of them, and an explanation of agents lands on top of that.
What’s actually holding people back?
Two things show up in Workday’s numbers, and the third shows up in the requests I get.
They’re not sure it’s allowed. 41% of employees who use AI say they deliberately conceal it. It was 49% the month before. Only 35% say their employer’s expectations for disclosing AI use are very or extremely clear, and 19% say there are no expectations at all.
The fixing eats the saving. One in three employees say they lose at least half the time AI saves them to reworking its output. Only 44% say they can rely on AI output without closely checking it.
They can’t see where it fits. The request I get more than almost any other is some version of “can you build me an agent?” Often the person has one job in mind, and an agent is simply the only word they’ve heard for it.
Start with one piece of work they can check
A common version of that request is a Word report that has to be filled in from figures in a spreadsheet. You don’t need an agent for that. Copilot in Word can read the workbook and fill in the table while you watch, and I recorded the whole thing with no cuts in under three minutes. I cover the detail in Copilot agent vs Copilot in Word.
Look at why that example works for someone who’s unsure. It’s their report and their figures, and they know what each number should be. If Copilot gets one wrong, they’ll spot it. Nothing happens in the background.
Drafting an email from Copilot Chat into Outlook has the same quality. It stops at a draft, so the person still reads it, changes what they want and sends it themselves.
Workday lands in the same place: help people use AI on work that’s repeatable, lower risk and easy to review.
Start with one piece of someone’s week they can check themselves, in a file they already have open.
A person who’s had that go well a few times has less reason to hide it and a better sense of what Copilot is good at. They’re also the one who can tell you which parts of their job an agent should take on.
When does an agent earn its place?
Don’t get me wrong, agents have their place. A viewer on my Word video described a proposal template their team fills in, drawing on what each person enters and on knowledge the organisation already holds. Lots of people use it, each prompts it differently, and the result takes longer and comes out worse. That job needs something built, either an agent or a skill that gives everyone the same method.
One person filling in one report can use Copilot in the app. Build when many people do the same job the same way, and put someone in charge of looking after it. I use four questions before building an agent to sort one from the other.
Four things to do before your next agent conversation
If you’re championing Copilot in your organisation or just in your team, this is where I’d put the effort.
- Write down what’s approved, role by role. Your finance analyst and your executive assistant need different examples, so give each role two or three, in the app they use every day.
- Spell out the rules. Workday’s list is a good checklist: approved uses, disclosure expectations, data boundaries and human review. If four in ten of your people are hiding their use, the rules aren’t clear enough yet.
- Show each team one small win. Pick a file they already work in, before anyone mentions agents.
- Collect the agent requests. Every “can you build me an agent?” tells you where someone is feeling pain. A job that repeats across lots of people, done the same way each time, goes on your shortlist.
If you want somewhere to start, sit down with someone who fills in the same report every month from figures that live somewhere else. Point Copilot at the source file with a forward slash, ask it to fill in the report, and then ask them what else in their week looks like that.
Frequently asked questions
Should we roll out Copilot before building agents? For most teams, yes. An agent has to be built and looked after, so start with what people already have. If you’re still working out which Copilot that is, start with Microsoft 365 Copilot vs Copilot Chat.
Why are employees hiding their AI use? Unclear rules are a big part of it. Concealment fell sharply in Germany, down 23 points to 39%, and Workday says more explicit norms, dialogue and local workplace conditions are likely making AI use easier to acknowledge.
Do these figures apply to Australia? The survey covers nearly 6,000 workers globally, and the published summary doesn’t break out Australia. Treat the numbers as a signal and run a short pulse check in your own team if you want local figures.
