More and more Dynamics 365 and Power Platform projects now include some kind of AI agent. And sooner or later, the same question comes up: do you build it in Microsoft Copilot Studio or Microsoft Foundry?
The two get mentioned together a lot – sometimes as if they’re interchangeable – but they’re not. They’re built for different purposes, different people, and different levels of complexity. Knowing which one fits your situation can save you a lot of rework later.
This guide walks through the key differences, shows where each platform makes sense, and helps you decide which one to use.
What is Microsoft Copilot Studio?
Copilot Studio is Microsoft’s low-code tool for building AI agents. It’s part of the Power Platform, so it plugs straight into Dynamics 365 (via Dataverse) and works natively with Power Automate, Teams, SharePoint, and the rest of Microsoft 365.
You build in a visual canvas – dragging and connecting blocks rather than writing code. That makes it accessible to business users, functional consultants, and low-code developers who don’t need a developer looking over their shoulder for every change.
Core capabilities include:
- Native connectors to Microsoft Graph, SharePoint, and Dataverse
- Agent Flow (Power Automate) integration for workflow orchestration
- API- and plugin-based tool calling
- Prompt configuration with grounding to enterprise content
- Over 1,000 Power Platform connectors out of the box
- Built-in test console and publishing to Teams, Outlook, and web
Enterprise features include:
- Entra ID single sign-on
- Microsoft 365 DLP and Purview compliance
- Content safety controls
- Conversation transcripts and analytics in Power Platform Admin Center
- Environment-scoped ALM with solutions
- Role-based governance and audit logging
- Tenant policies for controlled rollout
What is Microsoft Foundry?
Microsoft Foundry is Microsoft’s code-first platform for building AI agents and applications when you need full control over how everything works – the models, the prompts, the tools, the data. While Copilot Studio hides most of the technical complexity under the hood, Foundry puts it in front of you.
Core capabilities include:
- Access to 1,600+ models including GPT-4o, Llama 3, Mistral, and bring-your-own-model options
- Fine-tuning capabilities for domain-specific customisation
- Retrieval-Augmented Generation (RAG) using Azure AI Search and vector indexing
- Agent and prompt orchestration graphs
- Evaluation dashboards, traceability, and A/B testing
- SDKs, CLI, and VS Code extensions for reproducible development
- Support for AI frameworks including Semantic Kernel and the M365 Agent Framework
Enterprise features include:
- Connections to Azure Blob Storage, Data Lake, Fabric, SQL, and Cosmos DB
- Deployment to web apps, containers, AKS, and private endpoints
- Azure RBAC, managed identities, and Key Vault-backed secrets
- Private networking and VNet-isolated inference
- Monitoring with Azure Monitor and Application Insights
- Full LLMOps pipeline support with CI/CD and evaluation
Copilot Studio vs Microsoft Foundry: key differences
| Copilot Studio | Microsoft Foundry | |
|---|---|---|
| Target user | Makers, functional consultants, low-code devs | Pro developers, data scientists, solution architects |
| Interface | Visual, drag-and-drop canvas | Code-first (SDK, CLI, VS Code) |
| Model access | Microsoft-hosted LLMs via abstraction layer | 1,600+ models, fine-tuning, custom/BYOM |
| Tool integration | Power Automate, 1,000+ connectors | Azure Functions, Logic Apps, REST APIs |
| Data access | Microsoft Graph, SharePoint, Dataverse | Azure AI Search, Fabric, SQL, Cosmos DB |
| Voice/audio processing | Black box — speech-to-text only | Full access to raw audio streams |
| Knowledge base scale | ~500 documents, SharePoint, WWW, topic-based knowledge grouping | Enterprise-grade, large/complex knowledge bases |
| Security | Microsoft 365 DLP, Purview, Entra ID | Azure RBAC, VNets, Key Vault, Purview |
| Deployment | Teams, Outlook, Web, Mobile | Web apps, APIs, containers, edge |
| Monitoring | Power Platform Admin Center | Azure Monitor, Application Insights |
| Licensing | Often included in M365/Dynamics licences | Azure consumption-based |
| Time to first agent | Days | Weeks |
| Maintainability at scale | Harder | Stronger |
When to use Copilot Studio
Copilot Studio is the right choice when you need to move fast and your project stays within the Microsoft 365 world. Some concrete situations where it fits:
You need to extend a Dynamics 365 built-in copilot. Adding knowledge sources to Copilot in Dynamics 365 Sales or Customer Service, or building a custom agent that connects to the AI Hub – Copilot Studio is the natural way to do this. No extra licences, no code required.
You’re running a POC or MVP. If you need something working quickly and want to keep costs down, Copilot Studio is a solid starting point. It’s especially useful when you’re still validating an idea and don’t want to commit to a full build yet.
Your knowledge base is manageable in size. Copilot Studio handles knowledge bases of up to around 500 documents well – that’s where it performs reliably and stays easy to manage.
Your team includes non-developer stakeholders. If functional consultants or business stakeholders need to update and manage the agent after it’s live, Copilot Studio’s visual interface makes that doable without pulling a developer back in every time.
When to switch to Microsoft Foundry
Usually the decision to move to Foundry happens when you hit a wall in Copilot Studio. Here are the clearest signs it’s time:
You need a custom or fine-tuned AI model. This is the biggest one. If a client needs a model trained on their own data, fine-tuned for a specific domain, or built on something other than OpenAI – Copilot Studio can’t do it. Foundry is your only option.
You need to work with raw audio. Copilot Studio only receives voice input after it’s already been converted to text – you don’t get access to the actual audio. If your scenario needs to work with the audio stream itself (biometric verification in a contact centre, detecting caller characteristics, and so on), Copilot Studio has no way in. Foundry does.
Your knowledge base is large or complicated. At enterprise scale – thousands of technical documents, content in multiple languages, data spread across different systems – Copilot Studio starts to struggle. Foundry, combined with Azure AI Search and vector indexing, is built to handle that kind of volume.
You have strict security or compliance requirements. Private networking, isolated inference, Key Vault-managed secrets, full audit pipelines – that’s Foundry territory. If your client has data residency or security requirements that go beyond standard Microsoft 365 compliance, Foundry gives you the control to meet them.
How they work together: three Dynamics 365 scenarios
The key thing to understand is that Copilot Studio and Foundry don’t have to be an either/or choice. Microsoft built them to work together. The pattern most enterprise teams end up with is straightforward: Copilot Studio handles the conversation and the user-facing layer; Foundry does the heavy processing in the background.
Here’s what that looks like in three real Dynamics 365 situations:
Scenario 1: Contact Centre Intent Routing
A customer gets in touch – via chat, email, SMS, or phone. Copilot Studio picks up the conversation and manages the flow inside Dynamics 365 Customer Service. Simple, well-defined queries get resolved straight away using your knowledge base. For anything more complex – advanced reasoning, sentiment scoring, biometric verification – the request goes to a Foundry-hosted agent. The answer comes back to Copilot Studio, which either surfaces it to a human agent or closes the case on its own.Scenario 2: Sales Lead Qualification with a Custom Scoring Model
Your sales team uses the built-in Sales Qualification Agent in Dynamics 365. For standard lead research and recommendations, Copilot Studio handles everything natively. But if your business has its own lead scoring model – built from years of your own win/loss data – that model lives in Foundry. Copilot Studio calls it via API as part of the qualification workflow, combining out-of-the-box speed with your own logic.Scenario 3: Knowledge Base Management in Customer Service
The Knowledge Management Agent in Dynamics 365 Customer Service drafts knowledge articles automatically from resolved cases. For straightforward cases, Copilot Studio handles the drafting and puts it in front of an agent for approval. Where the knowledge base is large, multilingual, or pulling from systems outside Microsoft 365, Foundry’s retrieval pipeline and Azure AI Search handle the heavy lifting – keeping answers accurate even at scale. Copilot Studio shows the result in the workspace; the complexity stays out of sight.Cost considerations
| Copilot Studio | Microsoft Foundry | |
|---|---|---|
| Licensing | Often included with Microsoft 365 or Dynamics licences; pay-as-you-go message credits available | Pay-as-you-go via Azure subscription - no separate platform licence |
| Model usage | Message credits consumed per interaction | Billed per token, per model, and per compute unit |
| Storage | Dataverse, SharePoint (included in existing licences) | Azure Blob, Data Lake, SQL - billed separately |
| Scaling | Limited by Power Platform quotas and message credit pools | Fully scalable with Azure infrastructure |
| Maintenance | Managed by Microsoft | Partially user-managed (model updates, scaling config, evaluation pipelines) |
| Hidden costs to watch | Credit consumption can escalate quickly with autonomous agents | Compute costs for fine-tuning and high-volume inference add up fast |
A note on automation: neither tool is always the answer
Before you build anything, it’s worth asking a more basic question: does this actually need AI?
A lot of use cases that end up as AI agents could be solved more reliably – and more cheaply – with a Power Automate flow. If the process is well-defined, repeatable, and always needs to produce the same structured output, automation is the better fit. Adding an AI model introduces variability and cost without adding anything useful.
The best setups usually combine both: an AI agent to handle the conversation and figure out what the user needs, and automation to carry out the actual steps. AI where it earns its place; automation where reliability matters more.
Frequently Asked Questions
Can Copilot Studio connect to custom AI models?
No. Copilot Studio works with Microsoft-hosted models (currently GPT-4 and GPT-5 via abstraction layer) and does not support custom or fine-tuned models. For custom models, Microsoft Foundry is required.
Does Microsoft Foundry require coding?
Yes, in most practical scenarios. Foundry offers both a studio interface and a code-first experience, but meaningful customisation: fine-tuning, custom pipelines, advanced orchestration, requires developer involvement.
Is Microsoft Foundry available with Dynamics 365 licences?
No. Foundry runs on Azure and is billed on a consumption basis based on the underlying services used (Azure OpenAI, Azure AI Search, etc.). Copilot Studio, by contrast, is often included in existing Microsoft 365 or Dynamics licences.
Which is better for a contact centre AI agent?
It depends on what the agent needs to do. For intent recognition, case routing, and knowledge base responses, Copilot Studio handles standard scenarios well. For anything involving raw audio processing, biometric verification, or custom models, Foundry is necessary.
Can I start with Copilot Studio and move to Foundry later?
Yes, and this is often the recommended path. Build your POC in Copilot Studio, validate the use case, identify where the walls are – then bring in Foundry for the components that require it. More model intelligence in Copilot Studio does not mean more control; even with GPT-5, Copilot Studio is built for ease and speed, while Foundry is built for teams that need to own the full AI lifecycle.
What is the difference between Copilot Studio and Microsoft Copilot?
Microsoft Copilot (the built-in assistant in Dynamics 365 Sales, Customer Service, etc.) is a ready-made product you consume. Copilot Studio is the platform you use to build and extend your own agents – including customising those built-in copilots with your own knowledge, tools, and logic.
Summary: the decision framework
When choosing between Copilot Studio and Microsoft Foundry for a Dynamics 365 or Power Platform project, four questions determine the answer:
- Does the scenario require a custom or fine-tuned model? → Foundry
- Does it require working with raw audio or non-text data? → Foundry
- Is the knowledge base large, complex, or highly sensitive? → Foundry
- Is everything else standard? → Start with Copilot Studio
When in doubt, start with Copilot Studio. Prove the use case, identify the constraints, and graduate to Foundry, or a hybrid of both, when the business case justifies it.
