1.4. The Microsoft Agent Platform Landscape
💡 First Principle: Microsoft ships two agent platforms because there are two audiences with different control/effort trade-offs — and the platforms are designed to compose, not compete. Copilot Studio gives makers speed with managed guardrails; Microsoft Foundry gives developers control with code. The integrated solutions AB-620 tests are usually both at once.
Product-boundary questions are classic Microsoft exam territory: given a requirement, pick the right product — or recognize that the right answer is Copilot Studio consuming a Foundry capability. Misjudge the boundary and whole scenario chains fall apart, which is why this section earns a place in first principles.
| Dimension | Copilot Studio | Microsoft Foundry |
|---|---|---|
| Audience | Makers, low-code builders | Pro-code developers, data scientists |
| Lives in | Power Platform (SaaS) | Azure (subscription-based) |
| Build style | Graphical designer, natural language | Code, SDKs, APIs |
| Models | Managed defaults + Foundry catalog | Full model catalog, fine-tuning |
| Governance | Environments, DLP, admin center | Azure RBAC, subscriptions |
| Best for | Conversational agents on M365 channels, rapid delivery | Custom models, bespoke agents, deep Azure integration |
| Compose via | Foundry models in prompts, Foundry agents as connected agents | Expose agents for Copilot Studio to consume |
⚠️ Common Misconception: Foundry replaces Copilot Studio (or vice versa). They are complementary layers of one strategy — and the exam's title, integrated AI agent solutions, is a hint that the tested skill is making them work together.
