Microsoft Foundry vs Windows Agent Runtime vs Copilot Studio
Microsoft Foundry vs Windows Agent Runtime vs Copilot Studio (May 2026)
Microsoft is shipping three agent platforms in parallel — Foundry (cloud, code-first), Windows Agent Runtime (OS, dev preview), and Copilot Studio (low-code, business-user-first). They’re complements, not alternatives. Here’s how to navigate the stack, especially ahead of Microsoft Build 2026 (June 2–3, Fort Mason SF).
Last verified: May 31, 2026.
TL;DR
| Platform | What it is | Best for |
|---|---|---|
| Microsoft Foundry | Cloud platform for building/hosting/governing agents and models | Engineering teams building production agents at scale |
| Windows Agent Runtime | OS-level runtime on Windows desktops | ISVs building agents that integrate with desktop Windows apps |
| Copilot Studio | Low-code authoring tool with visual canvas | Business users and IT admins building workflow agents |
What each platform actually is
Microsoft Foundry (Azure AI Foundry)
Microsoft’s flagship cloud platform for AI agents and models. The unified successor to Azure AI Studio, Cognitive Services, Azure Machine Learning, and the agent-specific tools. It includes:
- Model catalog — hosted models from Anthropic (Claude Opus 4.8 added), OpenAI (GPT-5.5), Google (Gemini), Meta (Llama 4), Mistral, DeepSeek, Microsoft Phi.
- Foundry Agent Service — managed agent runtime with state, tool calling, memory, and persistence.
- Evaluation tools — agent evals, regression testing, safety scoring (Rampart Clarity for safety).
- Microsoft Graph grounding — pull data from Teams, SharePoint, OneDrive, Outlook.
- Governance — Purview integration, conditional access via Entra, audit logging.
- Deployment — push agents to Copilot, Teams, Windows, web embeds, custom apps.
Foundry is Microsoft’s answer to AWS Bedrock AgentCore plus Vertex AI Agent Builder combined, with deeper enterprise governance.
Windows Agent Runtime
The OS-level runtime on Windows desktops, reaching Insider preview in June 2026. It provides:
- Cross-app agent registration and discovery via Windows APIs.
- Sandboxed execution with enterprise policy (Intune + Purview).
- Standardized agent contracts (capability schemas, IO schemas).
- Companion: Windows Agent Store for distribution; Windows 365 for Agents for cloud runtime.
Initial scope: text agents on JSON/XML/PDF. Vision agents deferred to a later release. SDK is the Microsoft Agent Framework (open source, .NET + Python).
Copilot Studio
Microsoft’s low-code/no-code agent authoring product. Familiar to anyone who’s used Power Virtual Agents or Power Automate:
- Visual canvas for designing agent flows.
- Connector library — pre-built integrations with hundreds of SaaS apps and Microsoft 365 services.
- Multi-model picker — Anthropic Claude, OpenAI GPT-5.5, Microsoft default models.
- Deploys to Copilot, Teams, web, SMS, Slack, custom channels.
- Governance via Microsoft Entra and Purview, including Copilot Studio Agent Evaluations.
- Business-user friendly — no code required for many workflows.
Copilot Studio is Microsoft’s answer to Salesforce Agentforce Studio and ServiceNow’s agent builder, plus a successor to Power Virtual Agents.
Side-by-side
| Dimension | Microsoft Foundry | Windows Agent Runtime | Copilot Studio |
|---|---|---|---|
| Audience | Software engineers | App developers + ISVs | Business analysts, IT admins |
| Authoring style | Code-first (.NET, Python, TS) | Code-first (Agent Framework) | Visual canvas, low-code |
| Where agents run | Azure cloud, optionally hybrid | Windows desktop + Windows 365 for Agents | Microsoft Copilot, Teams, web |
| Model support | Catalog: Claude, GPT, Gemini, Llama, Mistral, DeepSeek, Phi | Pluggable: any | Multi-model picker, including Claude |
| Best for | Production agents at scale | Desktop-integrated agents | Workflow automation, customer support |
| Maturity | GA (Foundry rebrand of Azure AI Foundry) | Insider preview June 2026 | GA (successor to PVA) |
| Open source SDK | Various (Agent Framework, Semantic Kernel) | Agent Framework | No (visual canvas) |
| Governance | Purview + Entra | Intune + Purview | Purview + Copilot Studio admin |
| Multi-agent orchestration | First-class (Agent Service + Framework) | First-class (Agent Framework workflows) | Visual workflow chaining |
Where each one wins
Microsoft Foundry wins when…
- You’re an engineering team building production agents at scale.
- You need a managed agent runtime with state, memory, persistence, and observability.
- You want Microsoft Graph grounding for data-rich enterprise agents.
- You need to evaluate and regression-test agents before deploying.
- You want to host models alongside agents in one platform (model catalog + Foundry Agent Service).
Windows Agent Runtime wins when…
- You’re building agents that integrate with desktop Windows apps — Office, line-of-business apps, browser, file system.
- You need cross-app agent invocation within the OS.
- You’re targeting Windows enterprise fleets with Intune-managed desktops.
- You’re an ISV planning to ship through the Windows Agent Store.
Copilot Studio wins when…
- You’re a business user or IT admin who needs an agent but doesn’t write code.
- You’re building a workflow agent — customer support, IT helpdesk, sales sequencing, employee Q&A.
- You want deep Microsoft 365 + Teams integration out of the box.
- You want to deploy to Copilot, Teams, SMS, web from one canvas without rebuilding.
- You’re replacing or extending Power Virtual Agents or Power Automate with AI capability.
How they fit together
Microsoft is pushing a layered stack:
Microsoft 365 Copilot, Teams, Windows desktop (end-user surfaces)
▲
│ invocations
│
Copilot Studio (low-code) ── Foundry Agent Service (code-first)
▲
│ deployment
│
Microsoft Agent Framework (open SDK, .NET + Python)
▲
│
┌───────┴───────┐
│ │
Windows Agent Windows 365
Runtime for Agents
(local desktop) (cloud PCs)
│ │
└───────┬───────┘
▼
Model layer (Claude, GPT, Gemini, local) via Foundry catalog
Mental model:
- Foundry = build and govern
- Windows Agent Runtime + Windows 365 for Agents = execution surfaces (local + cloud)
- Copilot Studio = no-code authoring
- Microsoft 365 Copilot = end-user product
Comparison to non-Microsoft platforms
| Microsoft | Closest non-Microsoft equivalent |
|---|---|
| Microsoft Foundry | AWS Bedrock AgentCore + Vertex AI Agent Builder |
| Windows Agent Runtime | (no direct equivalent; closest: Apple Intelligence + App Intents, Android Gemini Intelligence) |
| Copilot Studio | Salesforce Agentforce Studio, ServiceNow agent builder, Camunda Processos |
| Microsoft 365 Copilot | Google Workspace Gemini, Notion AI, Slack agents |
Choosing for production
Most serious production stacks use 2–3 of these together:
| Production profile | Stack |
|---|---|
| Enterprise SaaS agent | Foundry (host) + Copilot Studio (front door for admins) + Microsoft 365 Copilot (user surface) |
| Desktop ISV agent | Agent Framework SDK + Windows Agent Runtime (target) + Windows Agent Store (distribution) + Foundry (model catalog) |
| Workflow automation | Copilot Studio (authoring) + Microsoft Graph (data) + Microsoft 365 Copilot (surface) |
| Codebase migration / engineering automation | Direct Claude Code with Dynamic Workflows OR Foundry-hosted Claude with custom orchestration |
What to watch at Build 2026 (June 2–3)
Three sessions to plan around:
- Nadella keynote (June 2 AM) — Windows-as-agent-platform framing, Foundry positioning.
- Agent Framework deep dives — SDK improvements, multi-agent workflow primitives.
- Copilot Studio sessions — multi-model picker maturity, governance updates, Studio Agent Evaluations.
Plus expected detail on Windows 365 for Agents pricing, agent identity model with Entra, and the Windows Agent Store opening.
Quick decision tree
Are you authoring agents?
├── With code (engineering team) → Foundry + Agent Framework SDK
├── With visual canvas (business user / IT admin) → Copilot Studio
└── Both? → Layer them: Foundry for hard logic, Studio for wrapping
Where do agents run?
├── Cloud (Azure) → Foundry Agent Service
├── User's Windows desktop → Windows Agent Runtime (June 2026 Insider preview)
├── Managed Cloud PC → Windows 365 for Agents
└── Multi-target → Agent Framework SDK (write once, deploy multiple)
Verdict
Don’t pick one — understand the layers. Microsoft Foundry is the engineering platform. Windows Agent Runtime is the desktop execution surface. Copilot Studio is the low-code authoring tool. Microsoft 365 Copilot is the end-user product. In serious production, you’ll use Foundry as your design + governance hub, Agent Framework SDK to write the agents, Windows Agent Runtime or Windows 365 for Agents as execution surfaces, Copilot Studio for low-code wrappers, and Microsoft 365 Copilot as the user-facing surface. That’s the stack Build 2026 is going to formalize, and it’s the most coherent enterprise agent story any vendor has shipped to date.
Sources: Microsoft Build 2026 conference page, Microsoft Tech Community FastTrack blog (Windows 365 for Agents, May 27, 2026), Windows News Build 2026 coverage, Microsoft Foundry product pages, Copilot Studio documentation, Microsoft Agent Framework GitHub (github.com/microsoft/agent-framework), Microsoft Copilot blog (Anthropic in Copilot Studio), PCMag Build 2026 preview, Help Net Security Windows 365 for Agents coverage (verified May 31, 2026).