Mission Objective
Welcome to your Digital Agency! By the end of this session, you will understand what Microsoft Foundry is and why it matters for business, know the 4 Core Building Blocks (Models, Knowledge, Agents, and Workflows), be able to describe an AI workflow in plain business terms, and complete your first "Job Ad" for a digital worker.
You've just been promoted. Congratulations! You're now the CEO of a brand-new Digital Agency. Your agency doesn't hire humans—it hires AI workers. These workers don't need coffee breaks, they don't call in sick, and they can process a thousand emails before you finish your morning coffee.
But here's the catch: AI workers are only as good as the instructions you give them. If you hire a brilliant analyst but tell them "just... do stuff," you'll get chaos. The same is true for AI. Today, we're going to learn how to write the perfect "Job Description" for your first digital employee.
This is Microsoft Foundry: Your AI Agency Headquarters. Think of Foundry as a factory for building AI applications. It's a unified platform where you can access smart "brains" for your AI (Models), connect your company's secret data (Knowledge Bases), create workers to do specific jobs (Agents), and make them work together (Workflows). The key insight is this: You don't write code. You design systems.
Key Takeaways:
- Understand what Microsoft Foundry is and why it matters for business
- Know the 4 Core Building Blocks: Models, Knowledge, Agents, and Workflows
- Be able to describe an AI workflow in plain business terms
- Complete your first "Job Ad" for a digital worker
The Gear List (Components)
Let's meet the building blocks of every AI application. Think of them as the essential departments in your agency:
Models: The Brains
A Model is the reasoning engine—the "brain" that processes information and generates responses. GPT-4o (The Professor) is expensive, slow, incredibly intelligent—use for strategy, creative writing, complex reasoning. Phi-3 (The Intern) is cheap, fast, good for simple tasks like summarizing or formatting. Foundry has a Model Leaderboard where you can compare models on quality, speed, and cost, plus a Model Router that automatically picks the best model for each task.
Knowledge: The Company Archive
Knowledge means giving your AI access to YOUR data. This is critical because, by default, AI models only know what's on the public internet. They don't know your company's price list, HR policies, or secret sales numbers. Without your data, AI will make things up (hallucinate). Create a Knowledge Base and connect it to SharePoint Online, OneLake/Microsoft Fabric, or uploaded files (PDFs, Excel sheets).
Agents: The Specialized Workers
An Agent is the component that does the work—a digital employee with a specific job. You build an agent by combining: a Model (the brain it uses), Instructions (its job description/System Prompt), Tools (its abilities like sending emails or searching the web), and Knowledge (the documents it can read). Key insight: An agent is NOT a chatbot. A chatbot just talks. An agent can TAKE ACTIONS: send an email, create a file, update a database.
Workflows: The Team Coordinator
A Workflow is how you make multiple agents work together. Imagine you have Agent A (generates YouTube titles), Agent B (generates descriptions), Agent C (edits the content). A workflow says: "First, run Agent A. Then, run Agent B. Then, pass both outputs to Agent C." Patterns include Sequential (step-by-step like a production line), Human-in-the-Loop (pauses for approval), and Group Chat (agents discuss dynamically).
How It All Works Together
Let's see a real example. Imagine a clothing company wants to use AI to research new market opportunities. A user asks: "What is a must-have apparel for the fall in the Pacific Northwest?"
Step-by-step: The Coordinator (Workflow) receives the request. Its logic says: "First, understand the intent." The Specialist (Agent A) is an "Intent Classifier" that determines this is a market research question. The Researcher (Agent B) is called—a "Market Researcher" agent.
Agent B uses GPT-4 (The Brain/Model) to process the request. Agent B queries the Knowledge Base to find product data from SharePoint and sales trends from Microsoft Fabric. Agent B uses a Web Search tool to find articles on current Pacific Northwest fashion trends. Agent B generates a report. The workflow presents it to the user.
This entire process—from question to answer—is orchestrated visually in Foundry. The Foundry Lifecycle includes: Discovery & Build (find models, build agents and workflows), Evaluate & Refine (test quality, review traces, fix errors), Secure & Govern (add safety filters, set policies), and Deploy & Operate (publish to Teams, monitor performance).
Key Points to Remember:
- Foundry is a factory—you don't code, you design systems
- The 4 Pillars: Models (Brain), Knowledge (Books), Agents (Workers), Workflows (SOPs)
- Agents take actions—they're not just chatbots
- Design before you build—write the "Job Ad" before touching the software
The Trail Map (The Foundry Lifecycle)
Field Notes: Write Your First "Job Ad"
You're going to write a Job Ad for your first AI agent using the Agent Strategy Worksheet.
- Job Title: What is the role? (e.g., "Customer FAQ Bot", "Product Price Scout")
- The Business Problem: What pain point does this solve? (e.g., "We answer the same 10 questions 100 times a day." or "Sales reps spend 2 hours/day looking up competitor prices.")
- Responsibilities: What will the agent do, step by step? (e.g., "1. Take a product name, 2. Search the web for competitor prices, 3. Return a comparison table")
- Qualifications: Does it need web access? Internal files? Math skills? (e.g., "Needs Web Search tool. No internal data needed.")
Ranger's Warnings (Common Pitfalls)
The Vague Instructions Trap
Never say "Find some information about competitors." Instead say "Find the retail price in USD from Amazon, Best Buy, and Walmart for each product on my list." AI workers are only as good as the instructions you give them.
The Hallucination Problem
Without your data, AI will make things up. Ask it "What is our return policy?" and it might invent one! Always connect Knowledge Bases for company-specific questions.
Confusing Chatbots with Agents
A chatbot just talks. An agent can TAKE ACTIONS: send an email, create a file, update a database. Design for action, not just conversation.
Pro Tips
Just like a CEO doesn't personally write every memo, you don't personally write every line of logic. You orchestrate. You say: "Agent A, research this topic. Agent B, write a summary. Agent C, send it to the client."
When you onboard a new employee, you give them the "Company Handbook." The Knowledge Base is that handbook for your AI.