Turn merged PRs into changelogs that write themselves

Changelogs are the first thing to rot on a fast-moving team. Someone scrolls through git log the day before a release, copies a few commit subjects, and calls it done - while half the actual work stays invisible. An AI changelog generator fixes this by working from the source of truth: the merged pull requests that make up a release, complete with their titles, descriptions, and linked issues.

FeatureFactory reads that history directly and drafts one clear entry per meaningful change, grouped into features, fixes, performance, and breaking changes. Terse or noisy commits get collapsed into readable lines, and PR and issue links are preserved so readers can dig in. It plugs into the same delivery data behind our pull request analytics, so the changelog is never a separate, hand-maintained artifact.

The result is a changelog that stays accurate because it is generated, not transcribed. You review and edit rather than author from scratch, which is the difference between a changelog that ships every release and one that quietly falls three versions behind.

Engineering and customers want different things from a release. Engineers want a dense, accurate log of every fix and breaking change. Customers want a readable story about the features they now have. Rewriting one into the other by hand is exactly the kind of busywork that gets skipped, so pick the audience and let the tone follow from the same PR data.

FeatureFactory generates both from a single pass over your Git history. Set the version, tag, or date range, choose the audience, and get either a terse changelog for CHANGELOG.md or a polished announcement grouped by theme. For the customer-facing path, the release notes generator handles the framing, while this generator keeps your internal log tight.

Because it is all built on the work you already merged, the changelog becomes a natural extension of measurement rather than a chore bolted on at the end. Read the manifesto for why we think shipping teams should measure and communicate what they ship from the same data.