Episode

While loops with tool calls

Podcast: Practical AI
Published: Oct 30, 2025
Duration seconds: 2685
Processing state: processed
Canonical source: https://share.transistor.fm/s/ca41b93c
Audio: https://pscrb.fm/rss/p/dts.podtrac.com/redirect.mp3/media.transistor.fm/ca41b93c/a54b2afe.mp3
JSON: /v1/public/podcasts/practical-ai/episodes/while-loops-with-tool-calls
Markdown: /podcast/practical-ai/while-loops-with-tool-calls.md

Actions

POST https://stenobird.com/v1/public/podcasts/practical-ai/episodes/while-loops-with-tool-calls/transcription-requests
Idempotently request low-priority transcript generation for this episode.
GET https://stenobird.com/podcast/practical-ai/while-loops-with-tool-calls.md
Read the agent-friendly Markdown representation of this episode resource.

Summary

The shift from prompt engineering to context engineering is enabling a new era of autonomous agents. This discussion explores how tool calling and 'while loops' allow models to iterate through tasks like humans do, rather than following rigid flowcharts.

Topics

Prompt Engineering
Context Engineering
AI Agents
Tool Calling
LLM Evaluation
Coding Agents
Structured Outputs
AI Development

Highlights

Main idea: Prompt engineering is evolving into context engineering, where the focus is on managing the environment and tools available to the model
Practical takeaway: Use a 'crawl, walk, run' approach to deployment, starting with small, scoped agents that can be easily validated
Failure mode: Autonomous 'while loops' in tool calling make systems harder to test and keep on the rails compared to static chains
Main idea: The rise of coding agents like Claude Code demonstrates how models can now handle complex, multi-step workflows with minimal human instruction
Practical takeaway: Success in AI products increasingly depends on the 'non-engineering taste' and domain expertise applied to the context provided to the model

Chapters

1:00 Introduction: Hosts Daniel Whitenack and Chris Benson welcome Jared Zoneraich, CEO of PromptLayer.
4:25 From Prompting to Context Engineering: A look at how the core architecture of LLM applications is shifting from simple input-output to complex context management.
8:00 The Evolution of AI Frameworks: Reflecting on how engineering approaches to LLMs have changed significantly over the last 18 months.
11:25 The Challenge of Evaluation: Discussing the difficulty of building heuristics and evaluations for increasingly autonomous systems.
17:50 The Power of Tool Calling: How native support for structured outputs and tool calling has replaced hacky prompting methods.
24:35 Testing Autonomous Loops: Analyzing the difficulty of unit testing the 'while loops' created by agents that iterate until a task is complete.
38:10 The Future of Coding Agents: How coding agents are transforming engineering workflows and democratizing AI product development.