Zed DeltaDB

Original Article ↗ Hacker News Discussion ↗

Zed’s new DeltaDB feature aims to record every code change and tie it to AI agent conversations, effectively creating a fine-grained, collaborative “local history” layer that sits alongside Git. Commenters are split: some see real value for multi-developer and agent-assisted workflows, easier rewind between commits, and future model training, while others worry about privacy, micromanagement, and scope creep at the expense of fixing basic editor stability and UX issues. The thread also reflects broader unease with VC-driven priorities and the growing entanglement of core developer tools with AI-centric workflows.

What DeltaDB Is Aiming To Do

  • Positioned as a layer on top of Git, not a Git replacement.
  • Tracks fine-grained “deltas” between commits, tying each change to the agent or conversation that produced it.
  • Built on Zed’s existing CRDT/multiplayer work; intended for collaborative and agentic workflows, including branching mid-run and joining live work.

Perceived Usefulness

  • Supporters compare it to JetBrains/VS Code “local history” but richer and shareable: easy rollback between commits, recovery from mistakes, and navigation from code back to the conversation that produced it.
  • Some find this helpful for debugging complex agent sessions or multi-person, multi-agent collaboration.
  • Others see it as niche: in years of work they’ve rarely needed this level of history and prefer frequent Git commits or tools like Jujutsu’s snapshots.

AI/Agent Workflows vs “Just an Editor”

  • Many like Zed’s ACP-based integration with external agents (Codex, Claude Code, OMP, OpenRouter) and see DeltaDB as a natural extension of “agent-native” workflows.
  • A sizable group wants Zed to focus on being a fast, solid editor (bug fixes, LSP stability, SSH/WSL, UI polish) rather than complex AI/VCS features.
  • Some users have abandoned Zed due to instability, performance issues, or missing “basic” features and feel DeltaDB is mis-prioritized.

Privacy, Surveillance, and Management Concerns

  • Strong discomfort about permanently logging every conversation and keystroke-like change.
  • Fears it will enable micromanagement (“bad prompt quality” as a performance metric) and incident retrospectives replaying agent chats.
  • Some worry it becomes de facto bossware or training data for future models; others note similar risks already exist with enterprise AI tools and spyware.

Git, JJ, and Other Comparisons

  • Multiple comments argue Git plus better UX, frequent commits, or Jujutsu snapshots already cover most needs.
  • Some suggest Zed could lean on JJ or integrate with existing tools instead of inventing new infrastructure.
  • Self-hosting is important to some; skepticism toward a VCS-like system that can’t be self-hosted.

Business & Strategy Concerns

  • Many tie DeltaDB and AI features to VC pressure: a plain editor is hard to monetize; AI workflows are sellable.
  • Worry that maintaining both an IDE and a VCS-scale system will dilute focus, especially given the current bug backlog and uneven platform support.