Replacing Pull Requests with Delta
summary
The traditional pull request workflow is outdated because AI agents now generate vast amounts of code that make manual diff reviews overwhelming and devoid of context. Delta replaces this batch-based process with a thread-based multiplayer environment where humans and agents collaborate in real-time. By preserving the entire evolution of a change through DeltaDB, the tool enables continuous engineering where implementation and review happen seamlessly in one place.
key points
- Delta replaces static pull requests with collaborative threads that allow teammates to enter a conversation with an agent and see the exact worktrees used during development.
- The system uses DeltaDB to track incremental edits and messages between commits, ensuring that reviewers have full context of why decisions were made rather than guessing from a final diff.
- Delta remains compatible with Git, allowing teams to use its real-time collaboration features while still pushing final checkpoints to GitHub for community access and version control.
who should read it
Software engineers and team leads who rely heavily on AI coding agents and find the traditional Git pull request process slow or cumbersome.
discussion skeptical
consensus
Commenters are generally skeptical about replacing pull requests with AI-driven chat histories, arguing that raw conversation logs are too noisy and 'slop' for effective review. While there is appreciation for Zed's innovation and the potential for faster feedback loops, most agree that reviewers prioritize polished, distilled results over the 'messy in-betweens' of AI interaction. The prevailing sentiment is that the core problem is not the UI, but the need for human-digestible logic and rigorous code verification.
top insight
The shift toward AI coding agents has decoupled the author's understanding from the final code, meaning traditional PRs—where the author explains their tradeoffs—are no longer sufficient. This suggests a need for new review tools because the human presenter may no longer possess the depth of knowledge required to justify every line of code produced by an agent.
Generated by an LLM from the article and thread. Check the sources before relying on it.