Intelligent Machines (Audio) IM 892: Pants-based Benchmarks - GitHub's Agent Revolution
56 snips
Oct 15, 2026 Martin Woodward, GitHub’s developer-relations VP and open-source advocate, explores how AI agents are transforming coding. They discuss agent-generated pull requests overwhelming maintainers, DevOps safeguards, multi-model Copilot workflows, and the emotional impact on professional developers. The conversation also covers personal software, AI memory and privacy, self-replicating models, Character AI’s shaky economics, and Nicolas Cage’s stand against AI waivers.
AI Snips
Chapters
Books
Transcript
Episode notes
Protect Open Source From Agentic Contribution Spam
- Maintainers should limit or temporarily disable pull requests when inexperienced contributors generate overwhelming low-quality submissions.
- Automated tests, security scans, and AI review can filter changes before human acceptance.
Personal Benchmarks Beat Model Leaderboards
- Leo Laporte’s personal benchmarks test whether models solve his actual workflows, not whether they top standardized leaderboards.
- Argon automated Slack-to-teleprompter ad transfers and produced a reliable Python script.
Prompt Agents With Goals And Verification
- Talk to agents conversationally rather than constructing elaborate prompts.
- State the goal, desired effort, and verification method, then ask the model to repeat its understanding before proceeding.






