Agentic AI Development with Agent Framework, MCP and .NET

Embrace the shift from deterministic microservices to agentic AI architectures by deploying AI agents as autonomous microservices that reason with goals and tools, using MCP for dynamic integration.

Trace the evolution from generative AI to agentic AI, shifting from single, stateless prompts to a coordinated, multi-agent system that orchestrates specialized agents to solve complex goals.

Learn how an AI agent functions as a digital employee that autonomously perceives, reasons, and acts to achieve goals using tools and APIs.

Explore the real-world challenges of AI agents, including non-determinism, security guardrails, and the high cost and latency of multi-step LLM reasoning, with emphasis on robustness and safe autonomy.

Explore the Microsoft Agent Framework and the .NET AI ecosystem, featuring enterprise-grade blocks, responsible AI by design, unified governance, and observability for scalable autonomous agents.

Master the three-tier .NET architecture—primitives, orchestration, and cloud—and embrace decoupling to upgrade cloud intelligence without rewriting core code, powered by Microsoft Foundry, Microsoft Agent Framework, and Microsoft Extensions AI libraries.

Dive into pillar three: Microsoft Foundry and Azure OpenAI, the secure hosting and governance layer for enterprise AI.

Architect enterprise agentic systems on the Microsoft stack by building a four-layer blueprint—language and runtime, orchestrator, hosting and observability, infrastructure—using C#, .NET, Aspire, and the Agent Framework.

Install dotnet 10 sdk and Azure CLI, provision a free Azure account, and prepare a dotnet console app within the Microsoft Foundry ecosystem using secure identities and az login.

Create a Microsoft Foundry project, configure a dedicated resource group and region, deploy the GPT-5 Mini model, and secure the endpoint, API key, and connection for .NET integration.

Prepare a hello agent .NET console app by installing Azure OpenAI, Azure Identity, and Microsoft Agents AI library, configuring endpoint and deployment via environment variables and authenticating with Azure CLI.

Learn to build your first AI agent in .NET using the Microsoft agent framework, defining a named, stateful agent with a system prompt and memory, testing with a Dev UI.

Compare synchronous non-streaming runAsync and streaming runStreamingAsync execution in the Microsoft Agent Framework to run agents. Learn how streaming lowers latency and supports observable metrics via OpenTelemetry.

Master multi-turn conversations by leveraging the agent session for context management, memory persistence, and continuous dialogue with the Azure OpenAI LLM in the Microsoft Agent Framework to power an enterprise assistant.

Learn to produce structured, deterministic JSON outputs with agents by defining strict C# records as data contracts and using the Microsoft agent framework to enforce schema and deserialize.

Explore tools in the Microsoft Agent Framework, where standard C-sharp methods become tools via the ai function factory, bridging enterprise resources and the llm.

Explore function tools with human-in-the-loop approvals to safely guard state-changing actions, using the Microsoft Agent Framework and ApprovalRequiredAI to pause and confirm refunds before execution.

Develop a minimal agent with function tools in an ASP.NET Core web API, enabling DevUI-driven visual testing, JSON payload auditing, and precise latency traces of tool calls and LLM reasoning.

Learn to add context and memory to agents by implementing short-term and long-term memory, managing chat history and sessions, and persisting state across conversations.

Explore creating conversational context with agent sessions and chat history, using AI context providers to bind user identity and department to each session.

Persist agent state across sessions using SerializeSessionAsync and DeserializeSessionAsync, storing the JSON session payload in Redis, Cosmos DB, MongoDB, or SQL databases.

Explore multi-agent workflows and orchestration with the Microsoft Agent Framework, contrasting traditional deterministic workflows with a stateful, non-deterministic, LLM-driven system.

Explore the anatomy of a modern agentic workflow, detailing the three core components—state, nodes, and edge—and how a manager llm routes tasks between research and summary nodes.

Develop a deterministic enterprise onboarding workflow using the Microsoft Agent Framework in .NET. Wire validation, enrichment, and audit nodes with a workflow builder and streaming execution.

Transform a workflow into a dynamic cognitive swarm by swapping functions for AI agents. Use the Microsoft Agent Framework for multi-service orchestration and encapsulation into a single runAsync agent.

Develop agents into the workflow by building an intelligent information technology helpdesk routing system with a swarm topology, routing tickets from a triage router to hardware and software specialists.

Explore enterprise multi-agent design patterns, including sequential handoff, concurrent, magentic sticky router, and group chat patterns, to orchestrate agentic workflows in .NET applications.

Develop agentic workflow patterns using the Agent Workflow Builder to generate sequential, concurrent, hand-off, and group-chat orchestrations, run by a centralized enterprise orchestrator with RunWorkflowAsync.

Showcases building a minimal agent and agentic workflows with triage, order, and refund agents in ASMED core, using DevUI and .NET Aspire for visual testing and auditing.

Examine the limitations of traditional reg, a retrieval augmented generation approach, and learn how agentic reg differs and works.

Explain agentic reg as an active researcher model where an LLM leads planning, reasoning, tool use, and reflection within an agentic loop, replacing traditional reg's passive pipelines and one-shot retrieval.

Compare agentic reg with traditional reg, revealing an active loop where the LLM orchestrates planning, searching, reflection, and self-correction, trading robustness for higher latency and cost.

Demonstrate how agentic RAG uses the active researcher loop to plan, reason, act, search, and reflect, updating strategies after each loop to determine the financial impact.