Is Copilot agentic AI?

We use Anthropic's published test from its Building effective agents post: a system is an agent when the model decides its own process and tool use, and a workflow when code fixes the path. Microsoft's own Copilot Studio overview, read on 7 October 2026, describes an agent in similar terms, as an assistant that uses tools to take action, reasoning through a request and deciding the best next step.

Copilot is also a family name. This page covers Microsoft 365 Copilot and Copilot Studio. GitHub Copilot is a separate product with its own agent features and its own documentation.

Microsoft's comparison page, dated 30 September 2026, describes two ways to build. A declarative agent adds instructions, knowledge and actions on top of Copilot's own orchestrator and models. The same page lists proactive interactions as not supported for declarative agents: they rely on interactions a user starts.

A custom engine agent brings its own orchestrator and models. Microsoft lists autonomy and proactive support among its parts, so it can start workflows and actions, make decisions and escalate. It usually needs hosting outside Microsoft 365, such as Azure, at extra cost, and can delegate tasks to other agents.

Copilot Studio is a low-code studio for agents and workflows. The overview page, updated on 1 October 2026, says the harness decides how an agent reasons. On the GitHub Copilot harness, an agent interprets the request, chooses tools and works through the steps without you scripting each path. On the standard harness, it matches a request to topics, which are conversation paths an author designs, and answers from knowledge when no topic fits.

Microsoft also says some agents can be given their own account so they work on tasks proactively, such as onboarding a new employee. Workflows include human-in-the-loop controls, and agent flows can include human review steps.

By Anthropic's test, a Copilot Studio agent on the GitHub Copilot harness is an agent, because it picks its own steps and tools. A custom engine agent with autonomy is an agent and can also start work without a prompt. A declarative agent chooses actions inside a conversation but waits for a person to begin, so it is an agent only within that turn.

A standard harness agent that follows authored topics behaves like a workflow, since the author fixed the path. Microsoft 365 Copilot answering a question from your Microsoft Graph data, with no agent involved, is an assistant rather than an agent.

Licensing changes what you can build. Microsoft's licensing page says the Copilot Studio trial lets you create and test agents but not publish them, and the Copilot Studio for Teams plan lacks generative orchestration. Check which harness each agent uses, which actions it can take, and who reviews them. Microsoft adds that custom engine agents leave compliance, responsible AI and security measures to you.

If your work lives in Microsoft 365, Copilot Studio is the natural place to start. Swfte is a separate platform and is not affiliated with Microsoft.

Swfte Studio builds agents and workflows that call any model reachable through Connect and pause at an approval step for a named person. SAML SSO and SCIM are in development, so check that before you plan a rollout across a large tenant.

• Swfte Studio
• Governance on the Swfte platform

Facts about other vendors and about the terms on this page were read on the pages below on 7 October 2026. Vendor plans, names and menus change, so check the vendor’s page before you rely on a detail.