Chapter 13 — Microsoft Foundry & Foundry Tools¶
Part III — AB-731 track: Leading AI Transformation
In 30 seconds¶
- The core idea: Microsoft Foundry is the platform for building and running custom AI solutions, and Foundry Tools provide the building blocks (models, search, vision). A leader maps use cases to these tools and matches a model to the business need.
- Why it matters: Foundry Tools are an explicit AB-731 objective.
- The exam angle: expect questions on Foundry Tools capabilities, matching a model to a need, and the benefits (scalability, security).
- Remember: Microsoft 365 Copilot is ready-made; Foundry is build-your-own for scenarios Copilot doesn't cover.
Exam map¶
Exam map — AB-731 · Domain 2: Identify benefits and capabilities of Foundry Tools
1. Key concepts¶
📖 Definition — Microsoft Foundry (Azure AI Foundry): Microsoft's platform for designing, customizing, and managing custom AI applications and agents — including access to a catalog of models and the tools to ground, evaluate, deploy, and monitor them.
📖 Definition — Foundry Tools: the capabilities within Foundry for building AI solutions, including the model catalog, Azure AI Search (for grounding/RAG), and vision (Azure AI Vision in Foundry Tools), among others.
| Foundry Tool | What it does | Example use |
|---|---|---|
| Model catalog | Choose from many pretrained models | Pick a model sized to the task and budget |
| Azure AI Search | Retrieval/grounding (RAG) over your data | Ground a custom app in your knowledge base |
| Azure AI Vision | Image understanding | Read text from scanned invoices |
| Azure AI (language, speech, etc.) | Language, speech, translation | Transcribe and analyze calls |
📌 Key concept: Foundry is where the build option (Chapter 12) lives. It's for custom, often customer-facing or specialized AI — not for the everyday productivity that Microsoft 365 Copilot already delivers.
Matching a model to a business need¶
Models differ in capability, cost, speed, and modality (text, image, audio). Choosing well means balancing these against the need.
🎯 Exam tip: "match an AI model to a business need" rewards fit, not maximalism. Pick the model that meets the requirement (quality, latency, modality) at acceptable cost — not simply the biggest or newest. This echoes token/cost thinking from Chapter 1.
2. How it works¶
flowchart LR
A["Business need"] --> B["Choose a model<br/>(model catalog)"]
B --> C["Ground it in your data<br/>(Azure AI Search / RAG)"]
C --> D["Add tools<br/>(vision, language, speech)"]
D --> E["Evaluate, deploy,<br/>monitor, scale"]🔍 How it works: Foundry provides the full lifecycle (Chapter 1) for a custom solution — select, ground, evaluate, deploy, monitor — with enterprise scalability and security built in.
🎯 Exam tip: Foundry's headline benefits are scalability and security — enterprise-grade infrastructure, governance, and the ability to grow from pilot to production.
3. In the real world¶
Scenario — beyond Copilot. A logistics company wants an AI service that reads scanned delivery notes and answers customer questions from its private shipment database — a customer-facing, specialized need that Microsoft 365 Copilot isn't built for. The team uses Microsoft Foundry: Azure AI Vision to extract text from the scans, a right-sized model from the catalog, and Azure AI Search to ground answers in the shipment data. They evaluate, deploy, and scale it on Foundry's secure infrastructure. This is the build path — chosen only because no ready-made Copilot capability fit.
4. Exam tips¶
🎯 Exam tip: Microsoft 365 Copilot vs Foundry — ready-made productivity vs custom-built AI. If a scenario is bespoke or customer-facing, think Foundry; if it's employee productivity, think Copilot.
🎯 Exam tip: Azure AI Search is the grounding/RAG tool for custom apps — the Foundry counterpart to Copilot's semantic index.
🎯 Exam tip: choose a model by fit (quality, cost, latency, modality), not by size.
5. Common pitfalls¶
⚠️ Pitfall: using Foundry to build what Microsoft 365 Copilot already does. Build custom only when ready-made doesn't fit (Chapter 12's build/buy/extend).
- Picking the biggest model by default: match to the need and budget.
- Skipping grounding: a custom app still needs RAG (Azure AI Search) to answer from your data.
- Ignoring the lifecycle: custom solutions need evaluation and monitoring, not just deployment.
6. Practice questions¶
1. A company needs a customer-facing AI app that extracts text from scanned documents and answers from a private database. Which platform fits?
- A. Microsoft 365 Copilot as-is
- B. Microsoft Foundry with Foundry Tools (vision + model + Azure AI Search)
- C. Outlook
- D. A saved prompt
Answer
Correct: B. A bespoke, customer-facing solution is the build path on Foundry, using vision, a chosen model, and Azure AI Search for grounding. Microsoft 365 Copilot is for employee productivity, not custom apps; C and D don't fit.
2. Which Foundry tool provides retrieval/grounding (RAG) over your data for a custom app?
- A. Azure AI Vision
- B. Azure AI Search
- C. PowerPoint
- D. Copilot Pages
Answer
Correct: B. Azure AI Search provides retrieval/grounding. Vision handles images; the others are unrelated.
3. How should a leader match a model to a business need?
- A. Always choose the largest, newest model
- B. Choose the model that meets the requirements (quality, latency, modality) at acceptable cost
- C. Choose the cheapest regardless of quality
- D. Let the model choose itself
Answer
Correct: B. Fit-for-purpose beats maximalism and beats blind cost-cutting. A over-spends; C risks quality; D isn't how it works.
4. What are the headline benefits of Microsoft Foundry?
- A. Scalability and security
- B. Free unlimited usage
- C. No need for responsible AI
- D. It replaces Microsoft 365 Copilot for all tasks
Answer
Correct: A. Foundry offers enterprise scalability and security. B is false (it's consumption-based); C is wrong (responsible AI always applies); D is false — Foundry is for custom builds, not everyday productivity.
Further reading¶
- Chapter 1 — Understanding Generative AI: model types, tokens, cost, and the ML lifecycle.
- Chapter 12 — Extending Copilot: build vs buy vs extend — where Foundry is the "build" option.
- Chapter 14 — Building the Business Case: Foundry Tools subscription/consumption models.
🔗 Source: Azure AI Foundry documentation (Microsoft Learn)