Copilot & Foundry: The Microsoft AI Stack for Small Business · Lesson 1 of 10
Why the Microsoft Stack Matters for SMB
Time: ~25 min reading + ~15 min worksheet Goal: Understand why M365 shops have a real home-field advantage with AI, learn the three-tier mental model that organizes everything else in this course, and meet the cast of characters.
---
If an SMB runs on Microsoft 365, they already have 80% of the infrastructure they need to do serious AI work. Their identity is in Entra (formerly Azure AD). Their documents are in SharePoint and OneDrive. Their conversations are in Teams and Outlook. Their admin plane is built. Their security posture exists.
That's not nothing. An AI deployment that lives inside Microsoft's ecosystem inherits all of it: single sign-on, DLP, sensitivity labels, audit logs, compliance certifications, BAA options, and the reflexive familiarity of every user.
Compare that to a bolted-on AI tool from a startup. Separate login. Separate permissioning. Separate audit trail. Separate training for the users. Separate vendor to vet. Separate contract to maintain. Every one of those is real friction.
For SMBs already on M365 Business Standard or Business Premium, the Microsoft AI stack is usually the path of least resistance — sometimes by a lot. This doesn't make it automatically the right choice (Claude, ChatGPT, local models, and purpose-built SaaS all have legitimate places), but it does make it the reasonable starting point for most M365 shops.
Your job in this course is to learn where the stack shines, where it disappoints, and when to reach for it.
Microsoft's AI offerings are sprawling. There are product names that change quarterly. There are overlapping tools. There are marketing decks that would confuse Microsoft's own salespeople.
Everything you need to know can be organized into three tiers. Memorize this. Use it in client conversations. It will save you.
Tier 1 — Consume: Microsoft 365 Copilot
You consume AI that Microsoft has built into products you already use. Outlook, Word, Excel, PowerPoint, Teams, OneNote. You don't configure much. You just turn it on for the right people and train them to use it well.
The pitch: "You already own the car. Now add power steering."
Tier 2 — Compose: Copilot Studio
You compose custom agents using a low-code visual tool. Topics, actions, connectors, and knowledge sources. Business users and citizen developers can do this. Deep integration with Power Platform. Publishes to Teams, SharePoint, websites, external channels.
The pitch: "You configure the agent to do your company's specific work, without writing real code."
Tier 3 — Customize: Microsoft Foundry
You customize production-grade AI applications in a pro-code environment. SDKs, a model catalog of 11,000+ models (including non-Microsoft ones), advanced RAG via Foundry IQ, enterprise observability, and on-prem/edge deployment via Foundry Local. This is where real software engineering meets AI.
The pitch: "You own and control the whole AI stack, end to end, with the flexibility a custom build needs."
The key insight
These tiers are not alternatives — they're complementary layers of the same stack. Microsoft explicitly designed them to work together. The most common real-world architecture for a well-served SMB uses all three:
- M365 Copilot for everyday productivity (Tier 1)
- Copilot Studio agents for specific departmental workflows (Tier 2)
- A Foundry-powered backend for anything that needs real scale, custom logic, or regulated-data handling (Tier 3)
When you hear someone ask "should I use Copilot Studio *or* Foundry?" — the right answer is usually "both, for different things." The question is which one goes where.
Beyond the three tiers, you'll hear these names. Know them.
Microsoft 365 Copilot
The in-app AI assistant across M365 apps. Sometimes called "Copilot" without a prefix, which causes endless confusion.
Copilot Chat
A free (for M365 users) chat interface at copilot.microsoft.com that works like ChatGPT, with optional grounding on your M365 data if you have the paid license. Not to be confused with the in-app Copilot.
Copilot Studio
The low-code platform for building agents. Part of the Power Platform family.
Microsoft Foundry
The pro-code AI platform. Formerly Azure AI Studio, then Azure AI Foundry, now just Foundry. Expect the name to drift again.
Foundry Local
Run Foundry models on your own hardware (including edge devices, on-prem servers). Keeps data off the public cloud. Big deal for regulated industries.
Foundry Agent Service
Managed runtime for production agents built in Foundry. Handles orchestration, safety, observability.
Foundry IQ
The RAG/grounding layer in Foundry. Works with Azure AI Search.
Azure AI Search
The vector search / indexing service. Powers most serious RAG deployments on Azure. Not an AI product per se — an infrastructure service AI solutions depend on.
Power Platform
The broader low-code family Copilot Studio lives in. Includes Power Automate (workflows), Power Apps (custom apps), Power BI (reporting), Dataverse (database), and Power Pages (websites).
Microsoft Purview
Microsoft's data governance platform. Increasingly central to the AI governance story — sensitivity labels, DLP, DSPM for AI, audit logs.
Connectors
Pre-built integrations (over 1,000 of them) between Power Platform and external systems. Salesforce, SAP, ServiceNow, Jira, QuickBooks, you name it. One of the biggest reasons to choose Copilot Studio.
Entra ID
Formerly Azure AD. Microsoft's identity service. Controls who can access what across all Microsoft services.
Microsoft slaps "Copilot" on a lot of products. This course covers the three tiers for building business AI solutions. We're not covering:
- GitHub Copilot — the coding assistant for developers
- Security Copilot — for SOC/security analysts
- Sales Copilot / Dynamics 365 Copilot — Dynamics-specific, narrow use case
- Windows Copilot — the OS-level assistant
- Copilot Pro (consumer) — the $20/month individual tier, not relevant for business deployments
If a client asks about one of these, you'll know what it is and whether it's the right tool. But we won't go deep on any of them here.
If you're consulting or running an IT services practice, this course is giving you one specific, marketable capability: the ability to walk into any M365 shop and architect an AI engagement across all three tiers with confidence.
That's a specific, saleable skill. Most IT consultants can speak vaguely to "enabling Copilot." A much smaller number can actually:
- Size a Copilot rollout and train the team
- Design and ship a Copilot Studio agent that solves a real problem
- Know when to escalate to a Foundry project and estimate what it would cost
- Stitch all three tiers together into one coherent client deliverable
If you can do all four, you can charge for it. That's the whole game here.
---
- M365 shops have a real home-field advantage with the Microsoft AI stack — existing identity, governance, and user familiarity are worth a lot.
- Three-tier model: Consume (M365 Copilot) → Compose (Copilot Studio) → Customize (Foundry). Memorize it.
- These tiers are complementary, not competitive. Most well-served SMBs use all three.
- Know the cast of characters — Foundry Local, Foundry IQ, Foundry Agent Service, Azure AI Search, Purview, Power Platform, connectors.
- The marketable skill is architecting across all three tiers, not just enabling one.
---
Next up: Module 2 — Tier 1: M365 Copilot. The part that comes out of the box.
---
Time: ~15 min Goal: Lock in the three-tier model and the cast of characters.
---
For each scenario, identify the tier that's primarily responsible: Consume (M365 Copilot), Compose (Copilot Studio), Customize (Foundry), or All Three.
- A CSR uses Copilot in Outlook to draft a polite reply to an unhappy customer. → ______________
- HR wants a chatbot in Teams that answers policy questions from the company handbook. → ______________
- A manufacturing client needs an AI system that analyzes 500K warranty claims, runs on their own VNet, and uses a fine-tuned model that isn't OpenAI. → ______________
- A healthcare practice wants to run all AI inference on hardware in their own office — no cloud calls. → ______________
- An ops manager asks Copilot in Excel to summarize a P&L and highlight anomalies. → ______________
- Sales leadership wants an agent that answers rep questions about products by searching a 300-PDF knowledge library in SharePoint, deployed to Teams. → ______________
- A large SMB wants to give everyone Copilot for productivity, build three internal-facing agents for HR/IT/Finance, AND build a customer-facing AI product that sits on their website. → ______________
Match the term to the one-sentence description. Don't peek.
| Term | | |---|---| | Foundry Local | | | Foundry IQ | | | Foundry Agent Service | | | Azure AI Search | | | Power Platform | | | Purview | | | Connectors | | | Entra ID |
Descriptions:
- A. The broader low-code family (Power Automate, Power Apps, Power BI, Dataverse, Power Pages) that Copilot Studio lives in
- B. The managed runtime for production agents built in Foundry
- C. Microsoft's identity service — formerly Azure AD
- D. Microsoft's data governance platform — sensitivity labels, DLP, audit logs
- E. The vector search / indexing service that powers most serious RAG on Azure
- F. Run Foundry models on your own hardware (on-prem, edge)
- G. The RAG/grounding layer in Foundry
- H. Pre-built integrations between Power Platform and external systems (1,000+ available)
You're on a 15-minute call with the owner of a 50-person professional services firm. They say: *"Someone told me we should do 'Copilot.' Someone else said 'Foundry.' What's the difference and what should I do?"*
Write a 90-second verbal response using the three-tier model. Use plain English. Don't name more than 3–4 products.
Your response: ```
```
Read it out loud. Under 90 seconds? If not, cut.
---
Exercise 1:
- Consume (M365 Copilot) — in-app assistant, no customization.
- Compose (Copilot Studio) — classic low-code agent use case with SharePoint knowledge.
- Customize (Foundry) — 500K records, custom VNet, non-OpenAI model = pro-code territory.
- Customize (Foundry Local specifically) — the edge/on-prem story.
- Consume (M365 Copilot) — in-app assistant in Excel.
- Compose (Copilot Studio) — 300 PDFs is within Copilot Studio's sweet spot.
- All Three — this is the complete-engagement scenario. M365 Copilot for everyone, Copilot Studio for internal agents, Foundry for the customer-facing product.
Exercise 2: Foundry Local → F, Foundry IQ → G, Foundry Agent Service → B, Azure AI Search → E, Power Platform → A, Purview → D, Connectors → H, Entra ID → C
Exercise 3 — sample response:
> "Microsoft has three layers, and you probably want to think about all three. The first is M365 Copilot — that's the thing built into Outlook, Word, Excel. You turn it on for your team and it makes everyone faster at writing emails, analyzing spreadsheets, summarizing meetings. That's the layer you start with. The second layer is Copilot Studio — it's where you build custom chatbots for your specific workflows, like an HR Q&A bot or a proposal helper. Most of the agents we'd build for you live there. The third layer is Foundry — that's where serious custom AI apps get built, usually for something customer-facing or where you need heavy-duty features. You probably don't need that layer yet. What I'd recommend is: start with the first layer, build one or two agents in the second layer within 90 days, and only talk about the third layer when we find a problem that genuinely needs it."
That's around 75 seconds. Confident, structured, doesn't oversell.
---
Check yourself: If the owner asks "okay, what does each of those cost?" — are you ready for Module 2? If you tripped on any of Exercise 2 acronyms, skim the cast of characters again before moving on.