A conversation with Ryan Booth, Engineering Manager at Juniper on AI development practices and related development tools.

Episode Description

Ryan Booth discusses his recent experiment building a complete application using AI assistance without writing code directly. He shares insights on managing AI development workflows, context management, testing practices, and practical tips for network engineers working with AI tools.

Key Topics Discussed

• Building applications using Claude 3.5 Sonnet through Cline (VS Code extension)
• Managing AI context and token limits in development
• Testing and validation strategies
• Frontend vs backend development experiences with AI
• Troubleshooting techniques when working with AI

Tools & Technologies Mentioned

• Claude 3.5 Sonnet
• Cline (VS Code extension)
• OpenRouter
• Ollama
• DeepSeek Coder
• LangChain
• LlamaIndex
• Ansible
• Redis

Key Points

• Break down development into focused tasks rather than trying to handle everything at once
• Maintain proper documentation and context files in directories
• Validate and test at each step rather than waiting until the end
• Use Git for granular version control of AI-generated code

Notable Quotes

• "I learned very early on when getting into the coding stuff that you can't overload it with information. You really have to kind of start just like you would a normal project. You have to build from the foundation up."
• "It's network automation is managing software at the end of the day. You're writing code that you have to rely on, that you have to test, that you have to validate."

Resources

Cline VS Code Extension: https://github.com/cline/cline
Claude AI: https://claude.ai
Claude AI Computer Use: https://www.anthropic.com/news/3-5-models-and-computer-use
OpenRouter: https://openrouter.ai

Episode Credits

Host: Kirk Byers
Guest: Ryan Booth
Recorded December 3, 2024

In this podcast, Kirk Byers and John Capobianco discuss the  impact of AI on network automation and engineering. They explore the significance of ChatGPT, the challenges of inference, and the concept of Retrieval-Augmented Generation (RAG). John shares insights on using LangChain for building AI applications, and the role of AI agents. The conversation emphasizes the importance of adapting to AI technologies and the potential for enhancing productivity in network engineering.

Takeaways

• ChatGPT marked a significant turning point in AI awareness.
• Retrieval-Augmented Generation (RAG) enhances AI capabilities.
• LangChain simplifies the integration of AI with network tools.
• AI agents can automate complex tasks in network management.
• Fine-tuning models can improve AI performance in specific domains.
• AI can significantly reduce the time needed for project development.

Chapters

00:00 - Introduction to AI and Network Automation

01:42 - The Impact of ChatGPT

05:50 - Understanding Hallucinations and Inference

09:53 - Retrieval-Augmented Generation (RAG) Explained

14:42 - Building with LangChain

18:37 - Exploring Models and Local LLMs

22:55 - Exploring Fine-Tuning and RAG Techniques

25:34 - Integrating AI with Network Data

29:34 - The Rise of AI Agents

34:28 - Modernizing Code

39:53 - Future Directions for Network Engineers

Reference Materials
Selector https://www.selector.ai/
John Capobianco YouTube Video on "Multi Agent AI for Network Automation" https://www.youtube.com/watch?v=8GwSIRGae10
LangChain https://www.langchain.com/
LlamaIndex https://www.llamaindex.ai/
Streamlit https://streamlit.io/