# Structuring Codebases for AI, Claude Code in GitHub, Scale Acquired! Granola Cafe, AI Rules and More MCP

Page: https://stenobird.com/podcast/you-ve-been-a-bad-agent-7334728/structuring-codebases-for-ai-claude-code-in-github-scale-acquired-granola-cafe-ai-rules-and-more-mcp
Text version: https://stenobird.com/podcast/you-ve-been-a-bad-agent-7334728/structuring-codebases-for-ai-claude-code-in-github-scale-acquired-granola-cafe-ai-rules-and-more-mcp.md
Podcast: [You've Been a Bad Agent](https://stenobird.com/podcast/you-ve-been-a-bad-agent-7334728)
Published: 2025-06-12T08:01:03+00:00
Episode link: https://share.transistor.fm/s/82181cce
Audio file: https://media.transistor.fm/82181cce/ad5f58a3.mp3
Processing state: processed
JSON: https://stenobird.com/v1/public/podcasts/you-ve-been-a-bad-agent-7334728/episodes/structuring-codebases-for-ai-claude-code-in-github-scale-acquired-granola-cafe-ai-rules-and-more-mcp
Duration seconds: 3826

## Resource

To leverage AI coding agents effectively, you must move beyond simple file access and focus on structuring codebases with clear boundaries and trace-based context. The discussion explores how tools like Claude Code and MCP servers can be optimized through better documentation and rule-based instructions.

## Highlights
- Main idea: Effective AI coding requires structuring codebases with clear boundaries and high-signal documentation to prevent model confusion
- Practical takeaway: Use compact representations of key code transactions (traces) to teach agents which functions and files are critical to core logic
- Failure mode: Models struggle with 'cleaning up' after themselves, necessitating strict rule files (like cursor-rules) to prevent unused variables and messy refactors
- Technical insight: Using URL-based secrets in MCP servers can bypass the complexities and broken implementations of standard OAuth flows in AI clients
- Tooling trend: The rise of subagents in tools like Shippie allows for parallel processing of complex coding tasks across different parts of a codebase

## Topics

Claude Code, Model Context Protocol, AI Coding Agents, Software Architecture, MCP Servers, Codebase Refactoring, Developer Tooling, Automated Migrations

## Chapters
- 1:00 — Tech Industry Drama & Meta's Scale Acquisition: A discussion on the implications of Meta's 49% stake in Scale AI and the strategic importance of data distribution.
- 11:00 — Codebase Structure for Parallelization: Exploring how separating testing frameworks, UI components, and logic allows for better parallel processing in AI-driven development.
- 15:00 — Automating Migrations with Claude Code: Using Claude Code and custom utility libraries to automate large-scale version migrations and codebase updates.
- 20:00 — Stateful AI Tools and Gitli: Discussing the emergence of stateful development tools and the utility of Gitli for managing developer workflows.
- 30:00 — Solving Authentication in MCP Servers: Analyzing the Supermemory MCP approach to using URL-based secrets to avoid broken OAuth and Bearer token implementations.
- 44:00 — Enforcing Strict Coding Standards via Rules: The necessity of using rule files to prevent models from introducing unused variables and unoptimized code.
- 53:00 — Context-Aware Agents and Trace-Based Learning: The future of coding agents: providing models with traces of key transactions to improve understanding of complex logic flows.

## Actions

- request_transcript: `POST https://stenobird.com/v1/public/podcasts/you-ve-been-a-bad-agent-7334728/episodes/structuring-codebases-for-ai-claude-code-in-github-scale-acquired-granola-cafe-ai-rules-and-more-mcp/transcription-requests` — Idempotently request low-priority transcript generation for this episode.
- read_markdown: `GET https://stenobird.com/podcast/you-ve-been-a-bad-agent-7334728/structuring-codebases-for-ai-claude-code-in-github-scale-acquired-granola-cafe-ai-rules-and-more-mcp.md` — Read the agent-friendly Markdown representation of this episode resource.

A page view does not enqueue transcription. Agents should invoke `request_transcript` explicitly when they need this episode processed.

## Transcript

Full transcripts are not published on public pages unless there is a clear rights basis.