Microsoft Copilot Studio is Microsoft’s low-code platform for building AI agents that answer from approved company knowledge and trigger real actions like workflows and approvals. It suits Australian organisations that want one consistent, governed agent for a repeatable process such as policy questions, helpdesk triage or job intake. The part most teams miss is that the platform rarely decides the result. The knowledge you ground it in and the scope you give it do.
Table of Contents
What is Microsoft Copilot Studio?
Microsoft Copilot Studio is Microsoft’s tool for building agents and workflows without heavy code. It is, in the words of the Copilot Studio overview on Microsoft Learn, a graphical, low-code studio for building and managing AI-powered agents and workflows. You connect them to your organisation’s data and publish them where staff already work, such as Microsoft Teams.
When Copilot Studio took over from Power Virtual Agents in November 2023, Microsoft’s announcement on Power Virtual Agents joining Copilot Studio told existing customers their bots would carry across. Our AI for Australian business guide shows where agents sit alongside assistants, Copilot and automation.
Five terms trip people up.
- An agent is an AI assistant built for one job and usually used through chat, such as answering policy questions or collecting a job request.
- A workflow is an automation that follows rule-based steps every time, with room for AI steps and a person’s input.
- Knowledge sources are the approved content an agent may answer from, for example current policy documents or a SharePoint library.
- Actions are the steps an agent takes beyond answering, such as logging a request, starting an approval or creating a record.
- Channels are where an agent lives and gets used, most often a one-to-one chat in Microsoft Teams.
The core capabilities are in this table.
| Capability | What it means in practice |
|---|---|
| Guided conversations | The agent asks the right questions in the right order |
| Knowledge grounded answers | Responses come from sources you choose and maintain |
| Actions and workflows | Steps run that create a real outcome |
| Channel deployment | The agent sits where staff already work, usually Teams |
| Governance and review | You control scope, access and how answers are checked |
What can you build with Microsoft Copilot Studio?
Copilot Studio works best when you build one agent for one clear job, with a question that keeps coming up and a next step that is the same each time. An agent asked to do everything usually gives vague answers and loses staff trust. A narrow first agent earns the right to a second.
The ABS release on business AI adoption put AI use at around 12 per cent of Australian businesses in 2024-25, with around 35 per cent of large businesses and 22 per cent of medium-sized ones using it. That leaves most Australian businesses at the start. For a first build, these use cases translate well. Each pairs a recurring question stream with a predictable action.
| Use case | Who it helps | Typical outcome |
|---|---|---|
| Staff policy assistant | HR, managers, new starters | Faster answers, more consistent guidance |
| Helpdesk triage agent | IT, operations | Better ticket quality, fewer follow-up questions |
| Job intake assistant | Service and trades teams | Complete requests, quicker scheduling |
| Sales support agent | Sales, admin | Faster replies from approved product information |
| Onboarding helper | People and culture | More consistent first-week steps |
If you can name a recurring question and the action that should follow it, you likely have a strong first agent. Our guide to what to build first with Copilot Studio agents works through that choice.
How do Copilot Studio agents, workflows and harnesses work?
Everything you build in Copilot Studio runs on a harness, the runtime that decides when to call the AI model, what to send it and which tools to use. There are three in Microsoft’s harnesses overview: the GitHub Copilot harness, the standard harness and the Copilot chat harness. Each runs different things and bills differently.
Despite the name, the GitHub Copilot harness is a Copilot Studio framework, not the GitHub Copilot service. Customer data isn’t sent to the GitHub Copilot service when agents run in Copilot Studio.
As at October 2026, the three compare like this.
| Harness | What it runs | Billing | Notes |
|---|---|---|---|
| GitHub Copilot harness | Agents and workflows for reasoning-heavy, multi-step work | Copilot Credits for all usage, from the moment you start building | Production-ready preview from June 2026 |
| Standard harness | Rule-based agents and agent flows | Copilot Credits after you publish, with some use included for licensed Microsoft Copilot users | Fully supported |
| Copilot chat harness | Agents that extend Microsoft Copilot Chat | Consumption-based or included in the Microsoft Copilot user licence | Publishes to internal teams |
Flows split the same way. Agent flows belong to the standard harness and feel like Power Automate inside Copilot Studio. Workflows belong to the GitHub Copilot harness. In its workflows overview, Microsoft calls them deterministic: they follow a rule-based path, with AI actions and steps that ask a person for input.
You pick the harness when you create an agent. To create a standard-harness agent, select “Other ways to build” on the Copilot Studio home page. An agent built on the GitHub Copilot harness or the standard harness can’t later be transferred to the other, according to the GitHub Copilot harness overview from Microsoft. One route across is a preview in Microsoft’s upgrade guide that copies a standard-harness agent onto the GitHub Copilot harness, leaving its knowledge sources behind. For a first agent on a predictable job, we usually start on the standard harness. It bills only after you publish.
Automation that sits outside an agent is often better in Power Automate, which our guide to AI in Power Automate covers.
Microsoft Copilot Studio vs Microsoft Copilot: which do you need?
Microsoft Copilot, formerly Microsoft 365 Copilot, is a per-user assistant inside Word, Outlook, Teams and the other Microsoft 365 apps. Copilot Studio builds agents and workflows that follow the same instructions and process for everyone who uses them. Match the tool to the outcome. Personal productivity points to Microsoft Copilot and a repeatable process points to Copilot Studio.
The rename, which also turned Microsoft 365 Copilot Chat into Microsoft Copilot Chat, is recorded on Microsoft’s Copilot requirements page.
| Comparison point | Microsoft Copilot Studio | Microsoft Copilot |
|---|---|---|
| Primary purpose | Agents and workflows for one process | Personal productivity across Microsoft 365 apps |
| Best for | Consistent answers, guided steps, structured outcomes | Drafting, summarising, individual help |
| Customisation | High, you control behaviour and scope | Low, tied to the app experience |
| Knowledge approach | You choose the sources and boundaries | Uses your existing files and permissions |
| Actions | Controlled actions through workflows | Mostly content tasks inside apps |
| How it is paid for | Copilot Credits | Per-user licence |
| Typical rollout | A department process with governance | Business-wide productivity uplift |
Microsoft Copilot tends to be enough when the user drives the task inside an app and the result stays there, such as drafting a document or summarising a meeting. Copilot Studio fits when the organisation needs the same behaviour every time, such as policy guidance, structured intake or helpdesk triage. Copilot Chat, the web-grounded version included with Microsoft 365 subscriptions, is a separate decision covered in our comparison of Copilot Chat and Microsoft 365 Copilot.
The two products also work together. Under Microsoft’s Copilot Studio billing rates, most employee-facing use of standard-harness agents in Microsoft 365 is included in a user’s Microsoft Copilot licence when the agent runs under that user’s identity. For agent flows, the inclusion covers only runs started by the “When an agent calls the flow” trigger.
Knowledge comes first with either product, as we saw on a Microsoft Copilot rollout for an Australian finance business. Staff wanted answers about policies and procedures, product and client information and compliance questions. The business’s knowledge base was not a sound foundation for Copilot. We restructured and cleaned the content, set clear owners and brought Copilot into Microsoft Teams where staff already worked. The client noticed fewer questions landing on one team, new starters finding answers for themselves and staff relying on what Copilot told them. The same order applies to Copilot Studio: structure the knowledge first, then build the agent on top. Our Microsoft Copilot integration work follows that order.
What does Microsoft Copilot Studio cost in Australia?
Copilot Studio bills in Copilot Credits. You buy them as monthly prepaid packs, through a one-year prepurchase plan or by pay-as-you-go with no commitment. What you pay depends on who uses an agent, where it runs, which harness it uses and what it does. Unused credits from a month’s capacity don’t roll over to the next month.
Older guides count usage in messages. Credits replaced messages on 1 September 2025 without changing the pack size or the pay-as-you-go rate, a change recorded on Microsoft Learn’s Copilot Studio licensing page. A prepaid pack is a tenant licence for 25,000 credits a month, the size given in Microsoft’s capacity pack guide.
Australian prices, excluding GST, are on Microsoft’s Copilot Studio pricing page for Australia.
Five things do most to move a Copilot Studio bill.
| Cost driver | What it changes |
|---|---|
| Who uses the agent | No credits for licensed Microsoft Copilot users’ answers from standard-harness agents in Microsoft 365 under their own sign-in |
| Where the agent runs | Channels outside Microsoft 365 are billed for every user |
| Which harness it uses | GitHub Copilot harness agents bill for all usage, including building |
| What the agent does | Microsoft Graph grounding and agent actions use more standard-harness credits than plain answers |
| How much it is used | Standard-harness credits are used per event, such as each answer, agent action or flow action |
For a standard-harness agent, estimate how many conversations and automated runs will draw credits each month and what the agent will do in each. In our builds those two estimates decide whether pay-as-you-go or a prepaid pack costs less.
Is your data safe in Copilot Studio?
Microsoft’s Copilot Studio application card says Microsoft doesn’t train foundation models on customer data. Australian environments store Copilot Studio data in Australia East and Australia Southeast, with exceptions. Two admin choices in particular can send model processing offshore: an admin setting for moving data across regions and an external model such as Anthropic’s Claude.
The storage exceptions are listed on Microsoft’s data locations page for Copilot Studio. App names and descriptions are stored globally. Agents’ activity history, including conversation logs, follows the region of the user’s Exchange mailbox. Microsoft Entra ID might store data globally. Channel integrations such as Microsoft Teams might transfer data outside Australia.
For Australian environments, Microsoft’s guidance on moving data across regions shows generative AI features running in region or in the United States when the Move data across regions setting is on. Processing moves to the United States only when a model isn’t deployed locally, local capacity is full or the local model has a reliability issue. Bing search runs in the United States.
External models can also send data offshore. In its model selection guide, Microsoft marks the Claude models available to Australian environments as cross-geo, meaning they might process data outside your region. Admins decide whether makers can add external models at all.
Permissions matter as much as location. Microsoft’s SharePoint knowledge guide is specific here: with SharePoint as its knowledge source, an agent only surfaces content the person asking has permission to open. Files uploaded from a computer straight into an agent work differently. Anyone with permission to use the agent can get answers from them.
Before an agent goes live, check five settings.
- Check whether the Move data across regions setting is on for the environment.
- Check which models makers may add, including external ones.
- Check whether knowledge comes from SharePoint or from uploaded files.
- Check which channels the agent publishes to. Cost and data location both depend on that choice.
- Check whether each tool runs on the user’s own sign-in or on the maker’s credentials. Makers can choose their own credentials unless an admin turns that option off.
One term needs care: enterprise data protection. Microsoft’s enterprise data protection page defines that term for users of Microsoft Copilot and Microsoft Copilot Chat. For agents used in Microsoft Copilot, the same page points you to each agent’s own privacy statement and terms of use.
Why do Copilot Studio agents underdeliver?
Most agents that disappoint fail for ordinary reasons, not technical ones. The scope creeps, the knowledge is messy, the agent answers without ever doing anything and nobody defined what happens when it can’t help. Each of those can be planned for before the build starts.
A Copilot Studio agent can only be as good as the knowledge it is grounded in. Point it at outdated, contradictory or scattered content and it will answer confidently and wrongly. Much of a business’s knowledge lives in SharePoint. Structuring that knowledge is often the real first project. Our SharePoint consulting work starts there.
Governance is the other half of the foundation. For an agent it is practical: clear limits on what it can do alone, monitoring of what it does and a record of its actions. Our guide to why AI pilots fail covers the wider pattern.
Frequently asked questions
Is Microsoft Copilot Studio the same as Power Virtual Agents?
They are related but not the same. Microsoft brought Power Virtual Agents into Copilot Studio in November 2023. Existing bots carried across. Copilot Studio has since added workflows and a choice of harnesses, so treat older Power Virtual Agents guides as background only and check them against Microsoft’s Copilot Studio documentation.
Do I need a Microsoft Copilot licence to use Copilot Studio?
No. Copilot Studio is paid for with Copilot Credits through prepaid packs, a prepurchase plan or pay-as-you-go. You can run agents for people who have no Microsoft Copilot seats once credits are available to the environment. A licence lowers the cost for the people who hold one. Answers they get from standard-harness agents inside Microsoft 365 under their own sign-in draw no credits. Agents on the GitHub Copilot harness bill credits whatever licence the user holds.
Is Copilot Studio free?
Standalone Copilot Studio isn’t free for live use. The Copilot Studio trial lets you build an agent and test it in the test chat, but you can’t publish it. Separately, Copilot Chat users get declarative agents grounded only in instructions and public websites at no extra cost. Those agents can’t reach your organisation’s data.
Can Copilot Studio use our SharePoint and internal documents as knowledge?
SharePoint sites and libraries are among the most common knowledge sources in the Australian builds we do. Point the agent at current, owned and accurate content instead of everything you have. An agent grounded on stale or contradictory documents answers confidently and wrongly.
Is our data used to train AI models?
Customer data in Copilot Studio isn’t used by Microsoft to train foundation models. Some models, including Anthropic’s, may process data outside Australia. Before launch, check which models your agents use and whether your admin allows data to move across regions.
Where should you start with Copilot Studio?
Start with one agent for one job that already repeats every week, built on knowledge someone owns. Check the data settings above before it goes live. One working agent teaches you more than a broad plan and gives you real usage figures to cost the next one.
If you want that first agent scoped and built properly, our AI agents team can take you from scope to build to adoption. Start with a free initial conversation about the process you want to improve and the knowledge it depends on.
