What is Copilot Studio? A practical guide for enterprises building agents in 2026
Copilot Studio is Microsoft's studio for building agents, workflows and apps on Microsoft 365 and Power Platform. What it contains, the three harnesses, licensing and when it is the right tool.

Microsoft Copilot Studio is a low-code studio for building AI agents, workflows and apps that run on Microsoft’s cloud, connect to your organization’s data and systems, and are published where people already work: Microsoft Teams, Microsoft 365 Copilot, websites, mobile apps and other channels. It sits on Power Platform, so it inherits its environments, connectors, data policies and admin center.
That definition is accurate and not very useful on its own. This guide explains what Copilot Studio actually contains as of November 2026, the decisions that shape every project, and when an enterprise should reach for something else. It will be updated as Microsoft changes the product, which happens often.
The building blocks
Copilot Studio now has four kinds of things you can build, and they can call each other:
- Agents. An AI assistant that follows instructions, draws on the knowledge sources you connect and uses tools to act. Agents can talk to employees or customers, and some can be given their own account so they work proactively inside a shared process.
- Workflows. Automations built in a drag-and-drop designer where each step can reason and act, with built-in testing and human review steps. A workflow can call an agent to complete one step.
- Agent flows. The established flow format, close to Power Automate but native to Copilot Studio. They run on their own or as a tool an agent calls, returning a result to it.
- Apps. App creation inside Copilot Studio, announced as generally available at Ignite 2026 and rolling out.
The practical rule: let the agent handle the parts that need judgment and language, and put the fixed sequence of steps in a flow or workflow, where it runs the same way every time.
The harness is the first architecture decision
Everything you build in Copilot Studio runs on a harness, the engine between your agent and the language model. Microsoft offers three, and the choice affects how the agent reasons, what it can do and how it is billed:
- Microsoft Copilot harness: reasoning-heavy, multistep work. The agent interprets the request, picks tools and works through the steps without a scripted path. Generally available since Ignite 2026.
- Customer Conversation harness: customer-facing voice and chat, with a primary agent and specialized sub-agents and a defined path to hand off to a human.
- Standard harness: rule-based agents and structured conversations built from topics, with generative answers from knowledge when a request falls outside them.
Billing follows the harness. The Standard harness has a per-feature rate card and starts charging after you publish. The Copilot harness charges for model tokens, tools and the harness itself, and starts the moment you begin building and testing. We covered what that change means for budgets.
What a Copilot Studio agent is made of
Channels
- Teams
- Microsoft 365 Copilot
- Websites and apps
- Voice
Agent, workflow or app
- Instructions
- Topics or goals
- Human review steps
Harness
- Microsoft Copilot
- Customer Conversation
- Standard
Decides reasoning, features and billing
Knowledge and tools
- SharePoint and files
- Dataverse
- Connectors and MCP
- Agent flows
Platform and governance
- Environments
- Data policies
- Solutions and pipelines
- Copilot Credits
Knowledge and tools
An agent is only as good as what it is grounded in. Copilot Studio can use SharePoint sites and files, uploaded documents, Dataverse tables, public websites, and enterprise data reached through connectors. Tools let the agent act: prebuilt and custom Power Platform connectors, MCP servers, agent flows and, on the Standard harness today, computer use through Windows 365 for Agents.
Two observations from practice. First, most poor answers trace back to the knowledge, not the model: duplicated, outdated or unowned documents in a SharePoint library produce confident, wrong answers. Who reviews the notes matters as much as which library you connect. Second, every connector an agent can call is a permission. Data policies decide which connectors can be combined in an environment, and they should be designed before the first agent reaches production.
How it is licensed, in one paragraph
Copilot Studio usage is measured in Copilot Credits. You can buy them as tenant-wide capacity packs, pay as you go through an Azure subscription, or commit up front through a pre-purchase plan. Makers need a Copilot Studio user license, which is free once the tenant has capacity, or authoring rights granted through a security group under pay-as-you-go. Users with Microsoft 365 Copilot licenses get many employee-facing agent interactions in Copilot Chat, Teams and SharePoint included, when the agent runs with their identity; autonomous agents and customer-facing agents still consume credits. Microsoft publishes a usage estimator, and the numbers change, so check the current rate card before any estimate.
Copilot Studio, Agent Builder or Foundry?
Microsoft offers three places to build agents, and they overlap:
- Agent Builder, inside Microsoft 365 Copilot, creates simple declarative agents over SharePoint content and Copilot connectors. It cannot author actions that call external services; for that, Microsoft tells you to copy the agent to Copilot Studio.
- Copilot Studio is the middle: low-code, with connectors, flows, multiple channels, environments and lifecycle management.
- Microsoft Foundry is the pro-code platform for custom models, complex orchestration and products built by engineering teams.
We have written about matching these tools to risk tiers. The short version: start with the smallest tool that does the job, and move up when you need actions, governance or scale.
When Copilot Studio is the right choice
It fits well when the organization already runs on Microsoft 365 and Power Platform, when the agent needs to act in business systems through connectors, when IT wants environments, data policies and an admin center it already knows, and when the builders are a mix of business makers and professional developers. Its weak spots are equally clear: costs that depend heavily on design, quality that depends on the state of your documents, and lifecycle management that has to be set up deliberately rather than inherited.
What to settle before the first agent
Before anyone builds: which environment the agent lives in and who owns it, which harness it uses, which knowledge sources it may read and who maintains them, which identity it acts with, what a month of use should cost, and how you will test it before each change. These are the questions that decide whether an agent reaches production or stays a demo.