Stenobird

Episode

Study Says AI Makes Developers Slower? F1 Movie Review, Coding and Testing for AI, Free Perplexity and Free Ideas

Podcast
You've Been a Bad Agent
Published
Jul 18, 2025
Duration seconds
4194
Processing state
processed
Canonical source
https://www.youtube.com/watch?v=9BvujIq5ZSg
Audio
https://media.transistor.fm/e3666ae9/8214d439.mp3
JSON
/v1/public/podcasts/you-ve-been-a-bad-agent-7334728/episodes/study-says-ai-makes-developers-slower-f1-movie-review-coding-and-testing-for-ai-free-perplexity-and-free-ideas
Markdown
/podcast/you-ve-been-a-bad-agent-7334728/study-says-ai-makes-developers-slower-f1-movie-review-coding-and-testing-for-ai-free-perplexity-and-free-ideas.md

Actions

  • POST https://stenobird.com/v1/public/podcasts/you-ve-been-a-bad-agent-7334728/episodes/study-says-ai-makes-developers-slower-f1-movie-review-coding-and-testing-for-ai-free-perplexity-and-free-ideas/transcription-requests
    Idempotently request low-priority transcript generation for this episode.
  • GET https://stenobird.com/podcast/you-ve-been-a-bad-agent-7334728/study-says-ai-makes-developers-slower-f1-movie-review-coding-and-testing-for-ai-free-perplexity-and-free-ideas.md
    Read the agent-friendly Markdown representation of this episode resource.

Summary

A deep dive into optimizing development workflows for AI agents, focusing on code structure and testing. The discussion explores how strict linting, co-located tests, and specific architectural patterns can prevent AI-driven regressions.

Topics

  • AI Coding Agents
  • Claude Code
  • Software Architecture
  • Cloudflare D1
  • Drizzle ORM
  • Unit Testing
  • Developer Productivity
  • LLM Prompt Engineering

Highlights

  • Main idea: AI coding efficiency is driven by code structure, specifically through co-location of tests and types
  • Practical takeaway: Use strict linting rules and type-only packages to create a feedback loop that agents can use to self-correct
  • Failure mode: Avoid non-standard naming conventions or complex directory structures that confuse LLM context windows
  • Practical takeaway: Implement 'claude.md' or similar documentation files at the directory level to provide localized instructions for agents
  • Main idea: The effectiveness of AI agents depends on their ability to run and verify their own work through automated test suites

Chapters

  1. 1:00 Personal Updates & Socials: A brief chat about video podcasting upgrades and the idea for a podcast Twitter account.
  2. 11:00 Architecting for AI Agents: Discussing the importance of co-locating tests and using strict linting to create verifiable code loops for agents.
  3. 17:00 Claude Code Experience: Reflections on using Claude Code for rapid framework iteration and the benefits of terminal-based tools.
  4. 22:00 The AI Speed Debate: Analyzing whether AI actually speeds up development or introduces new bottlenecks in the workflow.
  5. 32:00 Prompting Best Practices: Sharing strategies for working within existing codebases and managing agentic tasks.
  6. 43:00 The Evolution of Coding Agents: Comparing modern coding agents and their ability to manage complex, multi-step tasks like to-do lists.
  7. 1:09:00 Cloudflare D1 & Drizzle: Technical discussion on managing migrations and databases using Drizzle and Cloudflare Durable Objects.