Stenobird

Episode

Building Software With Thorsten Ball: Decoupling Code for Agents, Finding "Your Pain" and Germans Use the Winky Face

Podcast
You've Been a Bad Agent
Published
Sep 14, 2025
Duration seconds
4237
Processing state
processed
Canonical source
https://share.transistor.fm/s/f12074c7
Audio
https://media.transistor.fm/f12074c7/5607f850.mp3
JSON
/v1/public/podcasts/you-ve-been-a-bad-agent-7334728/episodes/building-software-with-thorsten-ball-decoupling-code-for-agents-finding-your-pain-and-germans-use-the-winky-face
Markdown
/podcast/you-ve-been-a-bad-agent-7334728/building-software-with-thorsten-ball-decoupling-code-for-agents-finding-your-pain-and-germans-use-the-winky-face.md

Actions

  • POST https://stenobird.com/v1/public/podcasts/you-ve-been-a-bad-agent-7334728/episodes/building-software-with-thorsten-ball-decoupling-code-for-agents-finding-your-pain-and-germans-use-the-winky-face/transcription-requests
    Idempotently request low-priority transcript generation for this episode.
  • GET https://stenobird.com/podcast/you-ve-been-a-bad-agent-7334728/building-software-with-thorsten-ball-decoupling-code-for-agents-finding-your-pain-and-germans-use-the-winky-face.md
    Read the agent-friendly Markdown representation of this episode resource.

Summary

Software engineer Thorsten Ball discusses the shift from traditional coding to managing AI agents. The conversation explores how engineers must evolve from writing logic to designing robust feedback loops and structured environments for LLMs.

Topics

  • AI Agents
  • Software Engineering
  • LLM Feedback Loops
  • Prompt Engineering
  • Developer Productivity
  • Code Structure
  • Automation
  • Technical Training

Highlights

  • Main idea: AI agents require structured environments and high-quality feedback loops rather than just better prompts
  • Practical takeaway: Engineers should focus on building tools that allow agents to validate their own outputs via browser or terminal visibility
  • Failure mode: Relying on AI to generate code without verifying execution leads to 'broken' software that is difficult to debug
  • Industry insight: The software industry lacks the formal, structured training programs found in sectors like finance
  • Technical shift: The role of the engineer is moving from writing implementation details to managing the 'triangle' of design, product, and AI orchestration

Chapters

  1. 1:00 AI-Generated Creative Process: A look at using 11 Labs and AI tools to create the podcast's musical jingle.
  2. 12:00 The New Standard of AI Content: Discussing how AI raises the entry-level expectations for web and software quality.
  3. 22:00 The Engineer's Journey: Thorsten reflects on his transition into programming and the role of personal interest in skill development.
  4. 38:00 The Training Gap in Tech: Comparing the lack of formal training in software engineering to the structured internal mobility at companies like Bloomberg.
  5. 43:00 The Evolving Engineering Triangle: How the relationship between designers, PMs, and engineers is being reshaped by AI capabilities.
  6. 54:00 Designing for Agentic Workflows: Using structured data formats like JSON to make codebases more accessible and navigable for AI agents.
  7. 59:00 The Power of Feedback Loops: Why the ability for an agent to 'see' a browser or terminal is more important than the raw intelligence of the model.