Copilot Studio brings agents, workflows, and apps together so makers can build solutions for business processes in one platform. It matters to makers and teams automating work across departments, especially when a process needs both AI reasoning and predictable steps. Microsoft Events demonstrates the updated maker experience and a natural-language agent creation flow in a YouTube session.
Copilot Studio is being positioned as a place to solve business-process problems, not just to create chatbots. Its building blocks serve different roles: agents reason through tasks, workflows run defined steps, and apps present information or controls to people using the solution. Combining those pieces can support a process from decision-making through execution and user interaction.
The maker experience has been rebuilt around a new agent designer. The session describes a layout where makers can see an agent's tools and capabilities and configure them inline. The workflow experience has also been redesigned, with a focus on speed, visual clarity, and the ability to combine deterministic automation with agent capabilities when a process needs them.
App building completes that set of capabilities. The session describes the apps capability as generally available and presents it as part of the same approach to process innovation. Makers can also use natural language through Cowork or an agentic coding harness to describe a business problem, rather than deciding in advance which parts must be agents and which must be workflows.
The platform is aimed at makers building solutions for how an organization operates: for example, automating a departmental process or coordinating work across a team. The session frames this as a broader audience for low-code and no-code making, while also showing that agentic coding tools can be another route into creating solutions. The starting point is a clearly described business problem.
The demonstration uses a financial-advisor scenario. A natural-language prompt asks Cowork to create an agent for research, analysis, and client communications. The resulting example uses a Morningstar MCP server for market information, an existing organizational email-template skill, workflows, and client holdings from a CRM system.
The example is useful because it illustrates a design choice, not just an authoring shortcut: the agent handles analysis while a defined workflow handles a consequential approval step. The session does not establish that every demonstrated connection or tool is available in every organization, so makers should validate the systems and resources they intend to use in their own environment.
The demonstration starts with a request in Cowork and then reviews the agent and related workflow. MCP, or Model Context Protocol, is the connection mechanism named for the external Morningstar server in this example. The agent also draws on an existing skill for branded email templates and uses a workflow to retrieve client-specific holdings from a CRM system.
Workflows remain valuable when a process needs predictable behavior. The example separates the agent’s analytical work from the send-or-reject decision, rather than relying on the model to make that decision. Makers can also add agents or models as workflow nodes, such as the demonstrated email-classification step for a shared mailbox.
The workflow designer supports different visual orientations, including top-to-bottom and left-to-right views. That can help makers inspect a process or present it to colleagues. The important design point is that workflow steps and agent reasoning can sit in the same overall solution without making every step depend on open-ended AI decisions.
Natural-language creation can assemble a starting point, but the example still involves reviewing agent instructions, tools, knowledge, and workflows. A maker should check that the selected data source is appropriate for the task and that an agent’s directions reflect the organization’s requirements. In the financial example, restricting market information to the named Morningstar MCP server is an explicit way to keep that task grounded in the intended source.
Not every process should be handed to an agent. The session contrasts agent reasoning with deterministic workflows, which it describes as faster, cheaper, and more reliable for the approval decision in the example. A practical design is to reserve agent steps for work that benefits from reasoning and use defined workflow logic for approvals and other predictable actions.
The session also describes governance and cost management as areas to consider when organizations scale their use of agents. It does not provide specific licensing prerequisites or a detailed rollout schedule for the agent and workflow experience. The app-building capability, by contrast, is described as generally available during the session.
The platform brings together agents, workflows, and apps. The session positions them as complementary parts of solutions for business processes.
Yes. Workflows can combine deterministic steps with agent or model nodes, such as classifying an incoming email before routing it through the rest of a workflow.
Connect the agent to a workflow that requests approval for the draft. In the demonstrated example, the workflow sends the email only after approval.
The session describes the apps capability as generally available.