Copilot Studio agents

    Copilot Studio agents, built and run in Microsoft 365.

    Copilot Studio agents answer from your sources, carry out steps in your systems and run in Teams or on your website. We build the first one with your business team.

    For business and IT teams in Microsoft 365 who want an agent in Teams or on their website and will maintain it themselves.

    • Copilot Studio
    • Agent flows
    • MCP
    • Teams
    • SharePoint
    • Dataverse
    In a workshop in a Bernese farmhouse a woman builds an agent in Copilot Studio on a laptop, the small robot agent beside her, a carpenter at work on the bench
    01

    Where Copilot Studio fits

    Copilot Studio fits when an agent should work inside Microsoft 365: with sources in SharePoint and Dataverse, actions through connectors and people who reach it in Teams or Copilot. After handover, your team maintains it in the authoring interface. When the interface, workflow or model needs more design freedom, a custom application on Microsoft Foundry, with OpenAI or Claude fits better.

    02

    What an agent is made of

    An agent is made of six building blocks. We build the first one together with your business team, on your real questions and documents, and hand it over so your team can develop it further on its own. Which tasks agents take on in sales, service, finance or HR, we describe under Agentic AI.

    • Knowledge with sourcesYou connect the agent to a SharePoint library, a Dataverse table or a website. It answers with the permissions of the person asking and links the passage in the document. How this becomes knowledge for the whole workforce is described under Company Brain.Result: every answer can be checked against its source.
    • Actions in your systemsThrough connectors, agent flows or an MCP server, the agent creates a record in Dataverse, starts an approval or writes to a line-of-business system. Each action gets its own permissions, and where it has consequences, the agent waits for a person.Result: the step is done and traceable in the history.
    • Triggers without a promptThe agent starts when an email arrives in the shared mailbox, a record appears in a list or an order comes in. It checks the details, creates the filing and asks the responsible person if something is missing.Result: a new order is recorded and complete when the team opens it.
    • ChannelsThe same agent runs in Teams, in Microsoft Copilot, on your website or on the phone. Employees ask it in chat, customers on the website, and it hands over to your team when it gets stuck.Result: you maintain one agent in one place, and it appears where the questions come up.
    • Identity and controlEvery new agent gets an Entra Agent ID. Your IT sets Conditional Access, DLP policies and connector permissions per agent and sees in the admin centre which agents run and what they consume.Result: agents are subject to the same rules and controls as the other identities in your tenant.
    • Tests and evaluationThe platform runs a test set of your real questions and expected answers, again after every change to sources or instructions. The analytics show which questions the agent answers and where it hands over to people.Result: your team sees before each release whether quality holds.
    03

    Sources, ownership, billing

    An agent needs three things. Sources that are right, with clean permissions. One role that owns what it answers and does. And billing that matches the use, with a cap per agent. Where it reads personal data, we record with your data protection officer what it processes and what is logged.

    04

    Before the agent opens to everyone

    The agent runs in your environment and is tested on your examples, including incomplete questions, missing sources and prohibited actions. The evaluation shows which questions it answers and where it hands over. One person on your team maintains the sources and can stop it. The pilot's credits give the cost frame before it is opened to everyone.

    Sound familiar? Tell us where you stand today.

    Contact

    What's new.

    As of: 30 September 2026

    • Since July 2026

      Every new agent gets an Entra Agent ID.

      Copilot Studio automatically creates a Microsoft Entra Agent ID for every new agent; opting out at environment level is no longer possible. Connector permissions, Conditional Access and DLP policies therefore apply per agent.

      Source: Microsoft Learn: What's new in Copilot Studio
    • Since June 2026

      New authoring experience, new primary models.

      The new authoring experience on the GitHub Copilot harness is available as a production-ready preview: Microsoft IQ for emails, calendar, files and Teams messages, reusable skills and memory per user. Claude Sonnet 5 and GPT-5.5 Chat are generally available as the primary model.

      Source: Microsoft Learn: What's new in Copilot Studio
    • Billing as of 14 September 2026

      Copilot Credits, included for licensed employees.

      Agents are billed in Copilot Credits, through prepaid packs or pay-as-you-go. Internal use by people with a Microsoft Copilot licence is included within fair-use limits, computer use excepted. In the Power Platform admin centre a monthly cap can be set per agent. This applies to the standard harness.

      Source: Microsoft Learn: Billing rates and management

    Agents from our catalogue.

    Sina answers enquiries in veterinary practices; the Abacus agent prepares HR data and payroll changes for approval.

    What IT teams ask before the first agent.

    Does every user need a Microsoft Copilot licence?

    No. An agent also works for people without a Copilot licence; their use is then billed in Copilot Credits, through a pack or pay-as-you-go. For employees with a Copilot licence, internal use is included. In the pilot we read off which option fits your group.

    What may the agent see and what not?

    It answers with the permissions of the person asking and acts with its own identity in Entra ID. Whatever is shared too widely, it will find. That is why we fix the sources, review sharing and guests and restrict connectors per agent before it starts.

    Can the agent act on its own, for example place an order?

    Yes, through actions: connectors, agent flows, MCP servers or computer use where no interface exists. Each action gets its own permissions. Where an action has consequences, the agent builds in an approval and waits until a person decides.

    Which model is inside the agent?

    Several primary models are available, and since June 2026 also Claude Sonnet 5 and GPT-5.5 Chat. We choose the model on the test set of your questions and repeat the test when Microsoft releases a new one. Claude models are off by default for Swiss tenants; your administrator has to enable them, and we record the terms that then apply beforehand.

    Can the same agent serve customers on the website?

    Yes, the same agent can be published on your website. Use by visitors is billed in Copilot Credits. What a website agent may answer, how it hands over to your team and who maintains it, we describe under Agentic Websites.

    How do we keep costs under control?

    Copilot Credits measure use. In the pilot we read the consumption per agent in the admin centre, set a monthly cap and compare it with Microsoft's estimator. So you know the frame before the agent is opened to everyone.

    Can our team change the agent itself?

    Yes, in the authoring interface: add sources, adjust instructions, approve actions. We hand over with one person who owns this, and with a routine in which every change first runs through the test set and only then goes live.

    How does an agent move from development into operation?

    We create the agent in a development environment, connect sources and actions and set up sign-in. Through Power Platform solutions it moves in a controlled way into test and production, and the test set runs before every step. For every connector we check beforehand which records it actually reaches.

    Agree on a useful next step

    Tell us about your current workflow, the systems involved and the result you need. In our first conversation, we identify what else we need to know to plan a useful next step.