The Zed team believes Git is no longer a good fit for the AI-agent era. Their argument: the truly important discussions happen before a commit is made, and just the code we push isn't enough anymore. Their DeltaDB stores every code edit, AI response, and intermediate state of the code
.
In my view, this won't catch on at scale, and here's why.

Instead of commits and code snapshots, we'd end up with a massive amount of metadata to sift through during review. With AI we already write more code and ship faster — now we'd also need to review how the developer + AI "thought" through the task. To me this blurs accountability somewhat, whereas with a committed change, the developer who commits it is fully responsible for it.

On the other hand, layering metadata on top of Git as-is seems like a good idea. Storing prompts, task descriptions, and similar context separately from the code itself looks like a solid solution.

That said, Zed isn't trying to replace Git — more like add an intermediate layer. We'll see how it plays out; for now it's a closed beta.