A Better Agentic AI Workflow? 🧠 My First Look At Delta (Zed)

Summary

The video is a sponsored first look at Delta by Zed, a collaborative environment for working with AI coding agents. Its central idea is to move from managing isolated context snapshots to continuous engineering: keeping conversations, code changes, and the reasons behind decisions connected over time.

How Delta Works

• Streams are the main workspace. A stream can begin with a conversation or idea, without a project attached; Git projects can be added later.
• Rather than using a conventional chat box, users write in the stream and send messages with Command/Ctrl+Enter. They can also highlight parts of an agent’s response or project files and attach comments or questions.
• Conversations, annotations, and code changes are linked, helping users understand not just what changed, but why.
• DeltaDB records fine-grained changes between Git commits. Git remains compatible and serves as a checkpoint system, while Delta aims to preserve the path and discussion between checkpoints.

Collaboration and Review

A stream can include multiple Git projects and be shared with others, who can join the same discussion and project context. Changes, comments, and activity sync in real time across the desktop app and web version. Users can also follow another participant’s current activity.

Shared participants can comment on code diffs and contribute to decisions. The video cautions that sharing may grant access to the repository, so permissions should be considered carefully.

Delta also includes a review and completion flow. It creates a sub-conversation and a guide summarizing the branch’s work and Git commits. Review approval is separate from publishing or merging: the demonstrated “land” step handles committing and pushing changes.

Demonstration and Practical Guidance

The presenter builds a “context bridge” skill: a reusable package of specialized instructions, scripts, and resources. The skill creates and updates a document that records decisions, their rationale, and links to relevant points in the Delta stream. Checkpoints and shareable links let someone jump directly to specific context in the discussion.

The presenter recommends trying Delta’s introductory walkthrough. He also mentions a separate Git tutorial for viewers who need help creating a Git repository.

Assessment

The presenter likes Delta’s collaborative approach and its ability to retain the origins of decisions. He notes that Delta provides more context, but users still need to make sure that context is useful.

For one-off agent tasks, a conventional coding tool may be sufficient. Delta may be more valuable when people need to track evolving decisions and hand work over through a shared, living workspace.

Main Speaker and Source

• Callum, who identifies himself as “Waterloos” (channel: Wanderloots), presents the review and demonstrations.
• Zed is identified as the sponsor and developer of Delta.