Catalog & Cocktails: The Honest, No-BS Data Podcast Building AI Agents with Lena Hall
Dec 17, 2025
Lena Hall, Senior Director of Developer Relations at Akamai and seasoned AI practitioner, joins the conversation to demystify AI agents. They explore the difference between coding agents and product agents, emphasizing their roles in natural-language processing. Lena shares practical advice on building agents, the importance of precise specifications, and the need for context management through knowledge graphs. The discussion also touches on emerging trends like context engineering and the future of coordinated agent networks. A deep dive into AI's exciting landscape!
AI Snips
Chapters
Transcript
Episode notes
Agents Are LLM Wrappers That Require Orchestration
- An agent is effectively a wrapper around an LLM that ties it to tools, deterministic code, or other agents in a reasoning loop.
- Building reliable agents is a software engineering problem requiring orchestration, inputs/outputs, and evaluation.
Prototype First, Then Build From Scratch
- Start small: use an agent builder or a popular framework like LangChain to prototype quickly and learn common patterns.
- For deeper learning, build an agent from scratch to understand prompts, temperature, context history, and tool integration.
Bigger Context Windows Don't Guarantee Better Outputs
- More context doesn't always improve LLM output; very long sessions can degrade quality and contradict earlier responses.
- You must compress, summarize, or restart sessions and manage where context matters to preserve output quality.