Class Details
Diese private Online-Class dauert vier Tage. Sie ist für Software Architects und Engineers, die Agentic Applications mit Microsoft Foundry und dem Microsoft Agent Framework bauen, orchestrieren und betreiben.
Sie beginnen mit den Foundry Essentials und den MCP Servern, die Ihre Agents aufrufen. Danach bauen Sie serverseitige Agents im Foundry Agent Service und Code-First Agents im Agent Framework und setzen sie zu Workflows und Multi-Agent Orchestrations zusammen. Zum Schluss hosten Sie Ihre Agents und machen sie mit Evaluation, Observability, Guardrails und Governance messbar und sicher.
Zielgruppe
- Software Architects und Engineers, die Agentic AI Solutions bauen
- AI Pro-Code Developers, die Agents in Python oder C# entwickeln
Voraussetzungen
- Python, C# und TypeScript
- Ein GitHub Account
- Eine Azure Subscription mit Zugriff auf Microsoft Foundry
- Azure Development Skills sind hilfreich
Lehrplan
Modul 1: Foundry Essentials für Agent Developers
- Microsoft Foundry: Projects & Resources
- Model Deployments & Model Router
- Foundry SDK & Microsoft.Extensions.AI
- Responses API: Conversations & Streaming
- Foundry Toolkit & Foundry MCP Server
- Context Engineering: Budgets, Compaction & Caching
Modul 2: MCP Server, Security & Hosting
- MCP Server: Tools, Resources & Prompts
- Transports & MCP Inspector Debugging
- MCP Authorization: OAuth 2.1 & PRM
- MCP auf Azure Functions hosten
- MCP Resilience: Retry & Circuit Breaker
- Foundry Toolbox: Managed MCP Endpoints
Modul 3: Foundry Agent Service: Tools, Knowledge & Memory
- Prompt Agents: Versions & Conversations
- Code Interpreter, File & Web Search
- Browser Automation & Computer Use
- Foundry IQ: Knowledge Bases & Retrieval
- Advanced RAG: Hybrid Search & Reranking
- Agent Memory: User & Procedural Memory
- Stateful Agentic Loops: Plan & Resume
- Voice Agents: Voice Live API
Modul 4: Agent Framework: Agents, Tools & Skills
- Chat Clients: Foundry, OpenAI & Claude
- AgentSession: State & Persistence
- Context Providers & Long-Term Memory
- Middleware: Agent, Function & Chat
- Function Tools, MCP & Approvals
- Agent Skills: SKILL.md & Progressive Disclosure
- Harness Agent: Plan, Execute & Compaction
Modul 5: Workflows & Multi-Agent Orchestration
- Workflows: Executors, Edges & Supersteps
- Checkpointing & Human-in-the-Loop Requests
- Sequential, Concurrent & Handoff Orchestration
- Group Chat & Magentic Orchestration
- Task Decomposition: Planner & Waves
- Resilient Orchestration: Quorum & Supervisors
- Declarative YAML & Functional Workflows
Modul 6: Hosting, Durable Agents & Protocols
- Hosted Agents: Containers & Agent Identity
- Foundry Routines: Schedules & Event Triggers
- azd AI Agent: Deploy & Monitor
- Durable Agents: Functions & Task Scheduler
- Self-Hosting: ASP.NET Core & Responses
- A2A Protocol: Agent Cards & Endpoints (optional)
- Enterprise A2A: Registry & Shared State (optional)
- AG-UI & CopilotKit: Web Front Ends (optional)
Modul 7: Evaluation, Observability & Optimization
- Tracing: OpenTelemetry & Application Insights
- Agent Evaluators: Task Adherence & Tools
- Evaluation Suites: Judges & Baselines
- Continuous Evaluation & AI Red Teaming
- Agent Optimizer: Instructions & Tool Tuning
- Performance & Kosten: Routing & Caching
- Incident Response: Replay & Postmortem
Modul 8: Security, Governance & Delivery
- Guardrails: Prompt Shields & Intervention Points
- Prompt Injection: Layered Defenses
- Entra Agent ID: On-Behalf-Of Flows
- Zero Trust: Tenant Isolation & Secrets
- CI/CD: GitHub Actions Quality Gates
- Agent Lifecycle: Versions, Quotas & Retirement
- Responsible AI: Fairness Probes & Audit
Organisatorisches
AI Harness: Bringen Sie Ihren eigenen mit

