The unit of review should be a narrative, not a diff.
Agents generate it faster than any human can read it. The volume isn't slowing down.
The "why" is in Linear, Notion, Slack. Never next to the diff. Reviewers have to reconstruct it from scratch.
A 40-file diff with no thread. So it sits open for days while the author waits.
We make the why travel with the change, giving you the full story.
The legacy session check ran on every request, hitting Redis even for unauthenticated routes. We replaced it with a token flow that short-circuits when no auth header is present.
The outline gets built from the diff the author already pushed. They shape it into something worth sharing.
Push your branch and open the PR the way you already do.
It groups the diff into sections, orders them so they build on each other, and writes a first pass of the prose between them.
Reviewers read top to bottom, like a chapter. They understand the change as a whole - and approve in a single pass.
They don't make the team understand the change. Every rubber-stamped diff is comprehension debt that compounds. We're not here to review code faster than you. We're here to make sure a human actually understood it. That's the only velocity that lasts.
No. The AI drafts it from your diff and linked context; you spend a few minutes editing.
No. Comments and approvals all sync two-way.
You're in the driver's seat. Editing a draft is seamless, and you publish only when it's right.
We're opening Narrate to a small number of teams. Leave your email and we'll be in touch as room opens up.