3.3.1. Microsoft Foundry Platform
💡 First Principle: Foundry is appropriate when you're building proprietary AI products, need custom or fine-tuned models, or require enterprise-grade AI governance that the platform layers don't provide. Most business scenarios don't need Foundry—but when you do, nothing else will suffice.
Microsoft Foundry (formerly Azure AI Foundry) provides:
| Component | Purpose | When Needed |
|---|---|---|
| Model Catalog | Over 10,000 models (OpenAI, Meta, Mistral, etc.) | Need specific models not in Copilot |
| Agent Service | Build and deploy AI agents at scale | Enterprise agent development |
| Foundry Tools | Pre-built AI capabilities (vision, speech, etc.) | Need specific AI capabilities |
| Foundry IQ | RAG implementation platform | Complex grounding scenarios |
| Control Plane | Governance, security, monitoring | Enterprise AI governance |
| Evaluations & content safety | Built-in content filters on model deployments plus quality and safety evaluations | Applying responsible AI before and after release |
| Foundry Local | Runs curated models on-device (PCs, edge hardware) | Offline, low-latency, or data-must-stay-local use cases |
How a Foundry build works: choose a model from the catalog → prepare data and fine-tune if needed → evaluate it and apply governance → deploy it. Deployment publishes the model as an API endpoint that applications call; beyond Azure-hosted endpoints, some models and AI services can also run in containers or on-device (Foundry Local) for edge scenarios. Foundry is pro-code—projects are typically led by developers and data scientists—whereas Copilot Studio is the low-code/no-code option for business makers.
Appropriate Foundry scenarios:
- Building a proprietary AI product for customers
- Fine-tuning models on specialized industry data
- Deploying specific models from the catalog
- Enterprise AI governance and compliance requirements
- Scenarios requiring models beyond OpenAI/GPT
⚠️ Exam Trap: The exam may describe a business chatbot scenario and offer Foundry as an option. Business chatbots typically don't need Foundry—Copilot Studio handles them. Foundry is for AI product development and specialized model requirements.
Reflection Question: A company wants to build a customer-facing AI product that will be sold to their clients. Should they use Copilot, Copilot Studio, or Foundry? Why?
