Advanced AutoGen: Custom Agents & Dynamic Tool Use

• Proficiency in Python programming
• Experience building with LLM-based applications
• Familiarity with function calling and multi-agent system design

Audience

• Senior developers
• Platform engineers
• AI architects

AutoGen is an open-source framework from Microsoft for building multi-agent applications that use LLMs, tools, memory, and user interaction.

This instructor-led, live training (online or onsite) is aimed at advanced-level developers and architects who wish to design and deploy deeply customized agents using AutoGen’s Python-based APIs, function-calling capabilities, and modular toolchains.

By the end of this training, participants will be able to:

• Develop custom agents with role-specific logic and tool routing.
• Build dynamic workflows using advanced function calling and context switching.
• Implement memory modules and planning frameworks within agent teams.
• Handle multi-agent error states and adaptive retry mechanisms.

Format of the Course

• Interactive lecture and discussion.
• Lots of exercises and practice.
• Hands-on implementation in a live-lab environment.

Course Customization Options

• To request a customized training for this course, please contact us to arrange.

Review of AutoGen Core Concepts

• Agent and group definitions
• Function calling and role chaining
• Limitations of built-in agents and where customization is needed

Building Custom Agents with Python

• Defining agent behavior using user_proxy and AssistantAgent subclasses
• Injecting role-specific logic and decision-making
• Creating reusable agent modules and mixins

Advanced Tool Integration and Routing

• Tool registration, binding, and invocation
• Conditionally routing inputs to specific tools
• Managing multi-step toolchains and composite actions

Planning and Context Management

• Designing task decomposers and intermediate planners
• Maintaining context across chained agents
• Implementing scoped memory for long-running sessions

Error Handling and Recovery Mechanisms

• Detecting and managing failed or incomplete interactions
• Agent-triggered retries and fallback logic
• Logging, debugging, and response validation

Multi-Agent Collaboration with Custom Roles

• Coordinating specialists within dynamic agent groups
• Orchestrating reasoning loops and cooperative workflows
• Role separation vs. role blending in task assignments

Real-World Deployment Strategies

• Optimizing for performance and cost (token use, caching)
• Embedding AutoGen workflows into web apps or pipelines
• Security, observability, and user feedback integration

Summary and Next Steps