Zed Delta, code review at a different level

It's been a year since Zed became my main editor. Fast, clean, zero bloat — I've said it before, I'll say it again.

But a few months ago, Zed shipped Delta, and that's another level.

What is Delta?

Delta is Zed's answer to a simple question: how do you review code that was written partly by agents? Not with a diff frozen on a commit, but with the full context: the conversation that produced the code, the decisions that were made, and comments anchored on the code as it evolves.

The agent is right there in the thread. When something looks wrong, you don't reconstruct the intention from a diff: you ask the agent to explain it or fix it. Everything stays connected to the code in real time.

My review workflow

On my side, I split the roles between two models:

• GPT 5.6 for global improvements and suggestions — refactoring, readability, overall consistency.
• GPT 6 Astra for security — analyzing flaws in my apps, especially launch-now.

And this is where it gets interesting. On launch-now, GPT 6 Astra surfaced more than 50 fixes to make. Flaws I hadn't seen, wrong security assumptions, poorly validated user input — the kind of stuff that ends up as a sleepless night.

What changes

The real change isn't the number of fixes. It's the loop: one agent writes, another agent reviews with a different lens, and I decide. Code review is no longer a step I dread — it's a pass that runs in parallel.

Zed for writing, Delta for reviewing, two specialized LLMs to cover the blind spots. The hardest part remains: making sure the proposed fixes serve the project, not just the checklist.

And honestly, for a tool that young, Delta already delivers.