4.6. Reflection Checkpoint

Key Takeaways

Before proceeding, ensure you can:

• Distinguish chatbots (single-turn) from agents (autonomous loops with tool access)
• Use Azure OpenAI Assistants with threads, tools (code_interpreter, file_search, function), and runs
• Recognize when run.status == "requires_action" signals the need to execute tool calls
• Implement Semantic Kernel plugins using the @kernel_function decorator
• Configure Autogen agents with appropriate human_input_mode settings
• Select the appropriate framework based on single-agent vs. multi-agent needs

Connecting Forward

Phase 5 shifts from generating content to analyzing it. The vision services you'll learn extract meaning from images—but you might combine them with agents (an agent that "sees" uploaded documents) or RAG systems (images enriched with extracted metadata for search).

Self-Check Questions

1. A company wants to automate code reviews where an AI analyzes pull requests, runs tests, and provides feedback. Should this be implemented as a chatbot, a single agent, or a multi-agent system? Why?

2. An Azure OpenAI Assistant run returns status: "requires_action" with tool_calls containing a function type. What must the application do before calling runs.submit_tool_outputs()?