Building Agentic LLM Workflows with AutoGen

This is the deck I have presented at Intel AI Dev Summit 2025.

Transcript

1. Building Agentic LLM Workflows with AutoGen Daron Yöndem Tech Lead for Application Innovation Microsoft http://daron.me @daronyondem
2. Welcome! • Why Agentic Systems Matter? • AutoGen: An Open Source Multi Agent Framework that can operate agents autonomously or collaborate with humans. • AG2 – Fork of AutoGen 0.2 • Microsoft AutoGen 0.4
3. Foundations of Agentic LLM Workflows • Autonomy • Collaboration • Adaptability
4. AutoGen Architecture • A unified conversation framework for multi-agent interactions • Assistant Agents + User Proxy Agents • As simple as it sounds.
5. Categories of Design Patterns for Agents • Reflection • Tool Use • Planning • Multi-agent collaboration
6. Reflection
7. Tool Use
8. Planning
9. Multi-Agent Collaboration
10. Single-Agent Task DEMO
11. From Single to Multi-Agent • Easy Agent Orchestration • State and Memory • Tool Integration • Error Handling & Control
12. Multi-Agent Problem Solving / Research Assistant DEMO
13. Advanced AutoGen Workflows • Extending Agent Abilities (Skills/Tools) • Incorporating Multiple Models • Custom Prompts & Behaviors • Memory and Context
14. Complex Multi-Agent Orchestration DEMO
15. Real-World Applications • Customer Support • Automation & Scripting • Knowledge Assistants
16. Challenges and Limitations • Complexity • Cost and Performance • Reliability
17. Further Learning AutoGen Discord - https://aka.ms/autogen-discord Magentic One • MultimodalWebSurfer, FileSurfer, and MagenticOneCoderAgent - https://aka.ms/magentic-one Code Samples used in this session • https://drn.fyi/codesamples
18. Thanks http://daron.me | @daronyondem Grab slides on http://decks.daron.me/