Claude Code - All Things Product Podcast with Teresa Torres & Petra Wille

Teresa Torres and Petra Wille discuss moving from Claude in the browser to Claude Code running on a local machine, where it can work directly with files and workflows. They cover practical setup such as project initialization and rule files, scoping access carefully, and starting with task management because feedback loops are fast and results easy to verify. The conversation highlights content retrieval across one's own archives and Teresa's use of Claude Code as a repeatable publishing pipeline, including metadata generation and fact-checking. The episode is candid about what breaks and how to debug it, making it useful for non-engineers exploring the tool.

01Key takeaways

• Start with a bounded, low-risk task such as task management to learn how Claude Code behaves.
• Scope the tool's file and account access deliberately, as you would for a new intern.
• Use initialization and persistent rule files so Claude keeps consistent context across sessions.
• When the tool gets stuck, ask it directly what it is doing to diagnose the failure.
• Build reusable commands for repeated tasks like retrieving past content or generating metadata.
• Verify AI-generated claims against source transcripts and research before publishing.

02Key sections

From browser to local machine

The hosts explain how Claude Code differs from the browser chat because it operates on files and folders on the user's computer. Petra shares her first attempts and the tool's tendency to treat everything as a code repository.

Setup and scoping access

Using initialization commands and rule files shapes Claude's behavior and context. The guests recommend treating the tool like an intern, granting limited access within a deliberately scoped environment.

Task management as an on-ramp

Task management is presented as a low-stakes first use case because outputs are easy to validate and feedback loops are quick. Practical tradeoffs such as exporting calendar data versus direct connections are discussed.

Content retrieval and publishing workflows

Searching one's own archive of posts, drafts, and transcripts is highlighted as a standout use. Teresa describes building reusable commands and a pipeline for podcast metadata, chapters, and fact-checking.

Debugging and advanced research systems

When Claude spirals, asking it what it is doing helps diagnose the problem. The episode closes with a detour into Zettelkasten-style research for improving rigor in thinking.

03From the post

“Listen to this episode on: Spotify | Apple Podcasts Petra finally carved out time to try Claude Code—and immediately ran into the “this tool thinks everything is a code repo” reality. In this episode, Petra and Teresa unpack what it’s like to move from Claude-in-the-browser to”

04Frameworks mentioned

Summary and takeaways written by PM Atlas; quotes are short excerpts. © the original author.