$Designing Agentic Workflows

This hub covers the fundamentals of designing, evaluating, and debugging agentic AI workflows.

Understanding Agents

What agents are and when they work well:

• Agentic AI can be less or more autonomous
• Agentic AI works better for well know, linear processes
• We should invest in AX (Agents Experience)

Design Patterns

Core patterns for building effective agents:

• Agentic design patterns
• Planning (Agentic Pattern) with code improves performance
• Reflection (Agentic Pattern) consistently outperforms direct generation on a variety of tasks
• Reflection (Agentic Pattern) can use external feedback from tools to improve output

Evaluation

How to measure agent performance:

• Agents can be evaluated objectively or subjectively
• "LLM as a judge" grading with a rubric gives more consistent results when evaluating Reflection (Agentic pattern)

Debugging & Improvement

Finding and fixing problems in agentic workflows:

• Agentic workflows are only as good as their weakest component + error analysis

Related

• [[ZETTELKASTEN/MAIN NOTES/HUB NOTES/$BuildingsoftwareproductswithAI\parallel$Building software products with AI]] - Broader context for AI-assisted development