Compare · Azure AI Foundry

Copilot Studio vs Power Virtual Agents: What changed and what it means for your agents

Power Virtual Agents was rebranded to Microsoft Copilot Studio in November 2023 — but this was far more than a name change. Under the hood, Microsoft rebuilt the platform around generative AI: GPT-4o grounding replaced rule-based LUIS intents, Semantic Kernel replaced the turn-based dialog engine, and autonomous agent capability arrived alongside the classic topic builder. If you built PVA bots, here is exactly what changed, what still works, and what your migration path looks like.

Quick Verdict

If you have existing PVA bots…

  • Your bots continue to work — no forced migration
  • Existing topics, entities, and flows are fully preserved
  • You can access Copilot Studio features from the same admin centre

New builds should use Copilot Studio…

  • Generative answers mean you no longer need a topic for every question
  • Azure AI Search integration enables RAG over your SharePoint and document stores
  • Semantic Kernel backend enables autonomous multi-step agents

The real capability gap is…

  • PVA could not answer questions outside its authored topics — Copilot Studio can
  • PVA was turn-based dialog; Copilot Studio supports autonomous task completion
  • PVA had no native Azure OpenAI connection; Copilot Studio does via AI Foundry

What actually changed

A technical breakdown of capability differences between Power Virtual Agents and its successor, Copilot Studio.

CapabilityCopilot Studio (2023+)Power Virtual Agents (legacy)
Model catalog breadth
GPT-4o grounding + model selection via Azure AI Foundry connection
Rule-based LUIS/CLU intents; no generative model integration
Enterprise security (Entra ID)
Entra ID SSO, Managed Identity, Conditional Access policies
Entra ID SSO supported — same auth model as Copilot Studio
Native M365 integration
Teams, SharePoint, Outlook, M365 Copilot as native deployment targets
Teams channel deployment supported; no SharePoint/Copilot integration
Managed Identity
Managed Identity to Azure AI Search, Azure OpenAI, and Power Platform connectors
Not available — connector auth via stored credentials or OAuth
Multi-agent orchestration
Autonomous agents with Semantic Kernel backend; multi-step planning
Turn-based dialog flows only; no autonomous agent capability
Low-code builder
Visual agent canvas with generative answers; topic fallback to GPT-4o
Full visual topic builder — mature, stable, well-documented
Compliance frameworks
Inherits full Azure/Power Platform compliance portfolio
Same compliance portfolio as Copilot Studio — no difference here
Generative answers (topic-free)
Answers questions without a matching topic by grounding on knowledge sources
Every answer requires a pre-authored topic — no generative fallback
Azure AI Search integration
Native connector to Azure AI Search for RAG over enterprise content
Not available natively; requires Power Automate flow workaround
Vendor lock-in risk
Power Platform ecosystem; Semantic Kernel plugins are portable
Power Platform ecosystem — same lock-in profile as Copilot Studio

Available Partial / workaround required Not available

Migration guidance for existing PVA deployments

Your PVA bots still work

Microsoft has not sunset Power Virtual Agents. Existing bots continue to run, and you can manage them from the same Power Platform admin centre. There is no forced migration deadline announced. However, new capability development has shifted entirely to Copilot Studio.

Should you migrate?

For bots that are working and meeting their KPIs, migration is not urgent. For bots that struggle with out-of-scope questions (a classic PVA limitation), migrating to Copilot Studio with generative answers enabled will immediately improve coverage without re-authoring topics.

What the migration actually involves

Copilot Studio reads PVA topic structures — most of the migration is in the admin centre and does not require re-authoring. The effort lies in enabling generative answers, connecting Azure AI Search knowledge sources, and testing that GPT-4o fallback answers are appropriately scoped.

The Azure AI Foundry connection

Copilot Studio connects to Azure AI Foundry via the Azure OpenAI connector in Power Platform. This means your Copilot Studio agents can invoke Azure OpenAI models, use Azure AI Search indexes, and call custom Prompt Flow endpoints — bringing the full Azure AI Foundry capability into a low-code surface.

Our recommendation

The rebrand from Power Virtual Agents to Copilot Studio represents a genuine architectural shift, not marketing. The underlying change — from LUIS-driven intent classification to GPT-4o generative answers with Semantic Kernel as the agent runtime — is material. It means Copilot Studio agents can answer questions that were never authored as topics, perform multi-step tasks autonomously, and ground their responses on live enterprise content via Azure AI Search.

For existing PVA deployments: do not rush to migrate for its own sake. Assess each bot against one question — does it regularly fail to answer user questions because there is no matching topic? If yes, migrating to Copilot Studio with generative answers will immediately improve user experience. If your PVA bot handles a narrow, well-defined use case (e.g., IT ticket submission, leave approval) where every user intent is already authored, the urgency is lower.

For new builds: there is no reason to use Power Virtual Agents. Copilot Studio is the current platform, it includes everything PVA had, and it adds GPT-4o grounding, Azure AI Search integration, and autonomous agent capability. All new Microsoft documentation and support resources are oriented toward Copilot Studio. Treat PVA as the legacy system it has effectively become.

Copilot Studio migration

Migrate your PVA bots to Copilot Studio — the right way

We audit your existing PVA deployments, prioritise migration candidates, and rebuild with GPT-4o grounding and Azure AI Search — giving you enterprise AI agents that actually answer questions.