Zed and Delta: collaborative AI coding workflows explained

Zed co-founder Nathan Sobo discusses the editor's Rust architecture and introduces Delta, an agent-focused development environment built around DeltaDB. He traces Zed's design to his work on Atom and Electron, arguing that browser overhead limited the performance and control he wanted. Zed's GPUI framework renders through the graphics processor. At recording, Sobo says it remains a crate inside the Zed repository under a permissive license.

DeltaDB is designed to track changes between Git commits and connect code with the conversations behind it. Sobo describes shared threads that capture prompts and agent output, with replicated work trees on participants' machines. Delta treats the conversation as a document where users can place a cursor and annotate responses. The system interoperates with Git, so teams can still open pull requests and run CI.

For readers considering an AI agent workflow on their own hardware, the execution distinction matters: Sobo says Delta's desktop app can use your computer, while browser users must use hosted runners. The speaker describes a planned launch with a centralized collaboration service at delta.dev. Conflict-free replication also does not eliminate semantic conflicts in code. The interview was recorded before launch, with beta access planned around publication; repository-wide annotation indexing was still under development.