Power Virtual Agents Is Now Copilot Studio: What Changed

Power Virtual Agents is now Microsoft Copilot Studio. Microsoft's own documentation URL for Power Virtual Agents redirects to the Copilot Studio overview, and the product lives at copilotstudio.microsoft.com.

If you're searching for Power Virtual Agents, that's the answer — but the rename undersells the change. PVA was a chatbot builder: you authored topics, matched user phrases to them, and designed conversation trees. Copilot Studio still does that, and it's now the smaller half of the product.

What it is today, in Microsoft's words: "a graphical, low-code studio for building and managing AI-powered agents and workflows." Three building blocks:

• Agents — AI assistants that handle conversations and complete tasks, reasoning about what to do next
• Workflows — drag-and-drop automations where each step can reason and act
• Agent flows — the established flow format, similar to Power Automate but native to Copilot Studio

What carried over from PVA

If you built on Power Virtual Agents, the concepts you know are still there:

• Topics — designed conversation paths with steps, questions, and conditions
• Natural language understanding matching a request to the best topic
• Channels — Teams, websites, mobile apps, and Microsoft 365 Copilot
• Connectors — prebuilt and custom, reaching your organization's data

Topic-based authoring now runs on what Microsoft calls the standard harness, described as suited to "rule-based agents and structured, repeatable conversations." That's PVA's model, still supported, still the right choice for a bounded FAQ or triage bot.

The harness choice is the real decision

The concept that didn't exist in PVA and now decides everything: a harness is the engine that carries out the work, and your choice affects how the agent reasons, how complex a task it can handle, and how it's billed.

The billing split matters more than the reasoning difference for most teams — but it isn't the clean licence-versus-usage divide it first looks like. Per Microsoft's own consumption table, GitHub Copilot harness agents are billed for all usage regardless of licence, and that covers building, testing and evaluating them, not just running them. Standard-harness agents are covered by a Microsoft 365 Copilot licence only when the user holds one and the agent is used in a Microsoft 365 channel — publish that same topic-based bot to a website, or serve a user without a licence, and it consumes Copilot Credits too. Workflows and prompts follow the same conditional rule: covered inside a standard-harness agent run by a licensed user, billed in every other case.

So the accurate version is conditional. A reasoning agent's cost always scales with usage. A topic-based bot's cost scales with usage in exactly the cases that fall outside Microsoft 365 channels or licensed users — which, for a public-facing support bot, may be most of them. Work out your channel mix and licence coverage before assuming a topic bot is a fixed cost.

That's the single most important thing to understand before building: the same requirement, built two ways, has two different cost shapes. Verify current details on Microsoft's own billing documentation rather than any comparison article — this is exactly the sort of thing that moves.

When to use it

Good fit:

• You're already in a Microsoft 365 estate, and identity, governance, and data access are solved
• The agent needs to live in Teams where people already work
• Your data is in systems Copilot Studio has connectors for
• Non-developers need to build and maintain the thing

Poor fit:

• You're outside the Microsoft estate — most of the value is the integration
• You need model choice or provider portability
• The agent's core job is something a general framework does better

That first point is the whole business case. Copilot Studio's advantage isn't the agent quality, it's that identity, permissions, and connectors are already handled. Outside a Microsoft estate, that advantage disappears and you're comparing it on features, where general frameworks compete well. Agentic AI frameworks covers the alternatives.

Practical notes

Start on the standard harness for bounded problems. If the job is answering thirty known questions and escalating everything else, a topic-based agent is usually cheaper and easier to debug than a reasoning one — though check the channel it will run in before treating that cost as fixed. The reasoning harness is for tasks you genuinely can't script.

Design escalation first, not last. The most important behaviour in any support agent is how fast it hands off to a person. Build the escalation path before the happy path.

Your content caps the quality. An agent grounded in inconsistent, out-of-date documentation produces fluent answers from inconsistent, out-of-date sources. Fix your top topics before rollout — this does more for resolution rate than any configuration. We cover that in conversational AI companies.

Governance is a feature here. Copilot Studio includes agent inventory, role-based access, and analytics. If you're consolidating scattered bots, that's often the actual reason to adopt it — see enterprise automation platform for the wider consolidation question.

If your recurring blocker is running AI setups people publish rather than building inside one vendor's estate, Taku mirrors a working AI setup into your own desktop workspace and runs it there. The free app library shows what's available to mirror. Taku is in Beta, and the Mac app is available now.

FAQ

What happened to Power Virtual Agents?

It became Microsoft Copilot Studio. Microsoft's Power Virtual Agents documentation URL redirects to the Copilot Studio overview, and the product is at copilotstudio.microsoft.com.

Is Copilot Studio the same as Power Virtual Agents?

It includes what PVA did — topics, natural language understanding, channels, connectors — and adds reasoning agents, workflows, and agent flows. Topic-based authoring now runs on the standard harness.

What is a harness in Copilot Studio?

The engine that carries out the work. Microsoft offers a GitHub Copilot harness for reasoning-heavy work — currently a production-ready preview — a standard harness for rule-based agents, and a Copilot chat harness for extending Microsoft 365 Copilot. The choice affects capability and billing.

How is Copilot Studio billed?

It depends on the harness and on how the agent is used. GitHub Copilot harness agents consume Copilot Credits for all usage regardless of licence, including building and testing. Standard-harness agents are covered by a Microsoft 365 Copilot licence when the user has one and the agent runs in a Microsoft 365 channel, and consume credits otherwise — so licensing covers a topic-based bot conditionally, not automatically. Check Microsoft's billing documentation for current details.

Should I use Copilot Studio or a general agent framework?

Copilot Studio if you're inside a Microsoft 365 estate and want identity, governance, and connectors handled. A general framework if you need model choice, portability, or you're outside that estate.

Key points

• Power Virtual Agents is now Copilot Studio; the old docs URL redirects there.
• Topic-based authoring survives as the standard harness and is still right for bounded bots.
• Harness choice drives both capability and billing, and licence coverage on the standard harness is conditional on Microsoft 365 channels and licensed users.
• The real advantage is Microsoft estate integration, not agent quality.
• Content quality caps results; fix your top topics before rollout.