“Everyone's in a race to replace GitHub": Zed launches Delta because agents made pull requests obsolete

Justy Okay, so Zed just announced Delta, and the framing is wild — 'everyone's in a race to replace GitHub' because pull requests are now obsolete.

Cody Yeah.

Justy Not slow, not awkward — obsolete. The claim is that agents generate code faster than humans can review it in the async pull-request model, so Zed's betting on threaded conversation instead.

Cody Okay, so let's unpack that. A pull request is fundamentally a human reading exercise — you get a diff, you read it, you approve or request changes. That model works fine when humans are writing code at human speed. But if an agent is generating hundreds of lines per minute, the async review queue breaks down. You can't keep up.

Justy Right.

Cody Delta's move is to collapse that latency by making review synchronous — the agent drafts, the human watches the thread, they interject in real time. It's less 'wait for the full PR to land, then read it all' and more 'code's going live as we talk about it.'

Justy Which sounds great for speed, but I keep hitting the same wall with this framing — who's actually doing agent-driven code generation at a scale where the PR model genuinely breaks? Like, Cursor exists, Devin exists, Claude Computer Use exists. Are teams really hitting a volume problem where they can't review changes fast enough?

Cody That's the right question.

Justy Because if you're not, then Delta is solving a problem you don't have yet. And if you are, then the question becomes — does a conversation thread actually give you better visibility into what the agent did, or does it just make it feel faster?

Cody The honest answer is probably both, depending on the team. If you're using an agent to generate a feature and you want real-time steering, yeah, threaded review catches issues early — the agent can correct course before it commits 500 lines down the wrong path. That's a real win.

Justy Mm-hm.

Cody But you lose something. A pull request is a structured gate — it has required approvals, it triggers CI, it creates a checkpoint before merge. A conversation thread is a lot more freeform. The real risk is that you're trading auditability and policy enforcement for speed, and you don't realize it until something lands in prod that shouldn't have.

Justy That's the part that keeps me skeptical. Like, Zed's bet is that agents plus humans in the same editor, talking in real time, is better than async review. And maybe it is for steering the agent. But for governance — for making sure that code meets policy, that it's been tested, that the right people signed off — that's all harder in a chat.

Cody Exactly. You can build those gates into Delta, obviously. But you have to build them intentionally. They don't come for free the way they do in a PR model.

Justy So the product story is solid — Zed's got a real editor, agent adoption is real, the friction of switching to a PR alternative is lower if you're already in Zed. But the claim that PRs are obsolete feels premature. You'd need to show that teams are actually hitting the review bottleneck AND that threaded review solves it better than, like, a faster async tool would.

Cody Right. Faster async is way easier than reimagining the whole review model.

Justy Exactly. And Delta's adoption path is probably pretty narrow right now — teams that are already using Zed, already using agents heavily, and already frustrated with GitHub. That's not zero, but it's not 'everyone's racing to replace GitHub' either.

Cody The other thing that's worth noting is that Cursor and others are building similar stuff — conversation-first coding, agent-as-coauthor, less emphasis on the PR gate. So it's not like Zed's alone in the thesis. But that also means the thesis is unproven. Nobody's actually shown that threaded review at scale is more reliable or faster than structured async review with good tooling.

Justy That's the bet you'd want to see resolved in, like, six months with a real customer story. Show me a team that switched from GitHub, that generates serious agent volume, and that's shipping faster with fewer bugs. If that exists, the narrative flips. But right now it's a product hypothesis with good positioning.

Cody I'd put maybe sixty-forty that Delta gains real traction with agent-forward teams by end of Q4. The editor integration is real, the UX story is solid, and the timing lines up with agents becoming table stakes. But sixty-forty assumes the verification model actually holds up, and we don't know that yet.

Justy That's fair. Like, I'm not skeptical that Zed's going to build a good product — they clearly are. I'm skeptical that they've actually identified a problem that's more urgent than 'we need better agents' or 'we need faster CI.' But if they're right about the bottleneck, it's a really smart move.

Cody And the nice thing is we don't have to guess for that long. If teams start shipping with Delta and their deployment frequency goes up without their error rate spiking, the thesis wins. If error rates spike, the thread model's not catching what structured review caught.

Justy Yeah. And we'll know pretty fast — agent-forward teams are loud about their process changes.

Cody They really are.

Justy So Delta's interesting, the timing's real, but I'm not sold yet that pull requests are actually dead. I think Zed's built something that might be better for a specific problem — high-volume agent review with live steering — but that's not the same as 'everyone's replacing GitHub.'

Cody Fair read. The product's real, the positioning's sharp, and the execution looks solid. The bet is whether the problem's as widespread as the headline claims.

Justy Ask me in October if we're seeing real adoption stories.