Changelog Interviews Building Zed's agentic editing
May 7, 2025
Nathan Sobo, a GitHub innovator, discusses the groundbreaking AI-native editor called Zed. He highlights the concept of agentic editing, enabling real-time collaboration at 120fps. The conversation delves into the competitive landscape of software tooling and the balance between innovation and quality assurance. Sobo also shares insights into integrating AI tech into coding environments, enhancing user experiences while emphasizing community engagement and feedback management in coding workflows. Tune in to discover the future of coding with Zed!
AI Snips
Chapters
Transcript
Episode notes
Building AI-Native Editor from Scratch
- Nathan Sobo highlights the strategic advantage of building Zed from the ground up with deep control over the editor technology.
- This foundational ownership positions Zed to uniquely integrate AI and elevate developer experiences over time.
Testing Stochastic AI Outputs
- Treat AI-powered tests as inherently flaky and set pass thresholds instead of expecting consistent pass/fail.
- Maintain rigorous unit tests and continuously improve tooling to manage this unpredictability effectively.
Rigor Needed Amid AI's Chaos
- Nathan emphasizes designing software with solid primitives and rigorous testing to avoid legacy code problems.
- AI introduces randomness, thus increasing the need for disciplined software design and error control.