# Release notes customers can read, from the PRs that shipped.

Release writer is an AI release notes generator for GitHub releases. Name a release and it reads the pull requests merged since the previous one, writes plain-language notes grouped into New, Improved and Fixed, saves them as a draft Notion page and posts a short summary. Every line traces to a pull request.

## Why do release notes get skipped?

Because writing them means rereading every pull request in the release and translating it into words a customer understands, right after the release is out and everyone has moved on. So the notes come late, list pull request titles nobody outside engineering can read, or never get written at all.

Without it:

- Release notes written days later, or never.
- Pull request titles pasted in for customers.
- Fixes nobody remembers to mention.
- Internal changes leaking into customer notes.

With Release writer:

- A draft for every release, in plain words.
- Grouped New, Improved and Fixed.
- Every line traced to a pull request.
- Internal changes counted, not listed.

## What goes into customer release notes?

Three groups customers can scan, plus a count of what was left out and why.

### New

Things customers can do now that they couldn't before, one plain sentence each. For example: Features, New pages.

### Improved

Changes to what customers already use: faster, clearer, or working in more cases. For example: Speed, Wording.

### Fixed

Bugs customers may have hit, described by what went wrong and what happens now. For example: Errors, Edge cases.

### Left out

Refactors, CI and dependency changes, counted in the summary and listed only in the sources. For example: Internal, Counted.

## How does an AI release notes generator work?

Every run follows the same five steps. GitHub is only read; the notes are written once, as a draft Notion page, then summarised in Slack.

1. **Finds the release before** (GitHub, reads): The newest older release of the same component, matched by its tag.
2. **Reads the changelog** (GitHub, reads): The release's list of pull requests is the source of truth.
3. **Reads each pull request** (GitHub, reads): Up to about 40. One it can't read is written from its changelog line, and marked.
4. **Saves the draft** (Notion, drafts): New, Improved and Fixed, under 300 words, with the source pull requests at the bottom.
5. **Posts the summary** (Slack, posts): The counts per group, anything left out, and the link.

When it runs:

- Webhook: A step in your release workflow calls its webhook with the new tag
- @mention: @release-writer write notes for web-v3.4.0
- Schedule: Fridays at 15:00, for the latest release of the web app

## From one sentence to a release notes draft.

Describe the job, let Arclyx draft the agent, test it once on a real release, create it, then read what lands in #releases. The full draft waits in Notion.

Example output in #releases, "Draft release notes · web v3.4.0 · New 1, Improved 2, Fixed 3" (placeholder companies):

- New · Exports: You can now export any report as a CSV file from its menu. From #466. Source: example-a.com/pull/466
- Improved · Search: Search keeps your filters when you go back, and loads faster in large workspaces. From #462 and #471. Source: example-a.com/pull/462
- Fixed · Schedules: Saving a weekday schedule failed in some time zones. This now works. From #474. Source: example-a.com/pull/474
- Fixed · Sign-in: Signing in from an email link no longer opens a blank page on some phones. From #479. Source: example-a.com/pull/479

Draft in Notion under Release notes · 7 pull requests read · 1 internal change left out

## One draft, four teams.

The same release notes answer a different question for each team that reads them.

- Product managers: A first draft of every release's notes, in customers' words.
- Engineering leads: Notes that trace each line to the pull request behind it.
- Support: What changed in each release, before customers ask.
- Marketing: The New items worth announcing, already in plain words.

## How do you write release notes with AI?

Start from what actually shipped: the pull requests in the release. Give an AI agent the release, let it read each pull request and rewrite the customer-visible changes in plain words, grouped New, Improved and Fixed, as a draft a person reviews. In Arclyx that's five steps, starting with one sentence.

1. **Describe the job**: One sentence: which repository, which Notion page the drafts go under, and where the summary goes.
2. **Connect GitHub and Notion**: The agent gets GitHub tools that list and read releases, and a Notion tool that only creates new pages.
3. **Name the parent page**: Every draft lands under that page, and the agent checks its title before writing.
4. **Try it on a past release**: Optional. A dry run holds the page and shows the summary, so you can read the notes first.
5. **Choose its trigger**: A webhook call from your release workflow, or @mention it with a tag.

The sentence that built it (Webhook, When a release ships):

> When a release is published in our GitHub repository, write customer-facing release notes from the PRs merged since the previous release of the same component, grouped into New, Improved and Fixed, in plain words. Save them as a draft Notion page under Release notes and post a short summary with the link to #releases.

### Or start from one of these sentences

- Weekly roundup: "Every Friday at 3, write one set of customer release notes for every release this week, grouped by product, as a draft Notion page."
- Support briefing: "When a release is published, list the fixes in it that customers may have reported, in plain words, and post them to #support."
- Announcement draft: "When a release is published, pick the New items worth announcing and draft a short changelog post in Notion."
- Internal notes: "When a release is published, write internal notes for the team: every pull request in the release, one line each, posted to #eng."

## AI release notes vs GitHub's generated notes.

Writing them by hand and GitHub's generated notes each cover part of the job. Here is what each one does, and doesn't, do.

| What you need | Written by hand | GitHub generated notes | Release writer |
| --- | --- | --- | --- |
| Written for | Whoever writes them | Developers: pull request titles | Customers, in plain words |
| Grouping | Whatever fits | By labels you configure | New, Improved and Fixed |
| Traced to pull requests | Rarely | Yes, the list is the pull requests | Yes, sources listed at the bottom |
| Internal changes | Judged by hand | Listed unless excluded by label | Left out and counted |
| Where it lands | A doc | The GitHub release | A draft Notion page and one Slack message |

Source: [GitHub docs: automatically generated release notes](https://docs.github.com/en/repositories/releasing-projects-on-github/automatically-generated-release-notes), checked 24 October 2026.

## Is it safe to give an AI agent your releases and Notion?

It reads your releases and pull requests, creates one draft page under the Notion page you choose and writes one message. Every run is recorded and has a spend ceiling by default.

### Every run is recorded

- Every step, in order: Every run records the tools it called, with a short preview of each result. Only the person who started a run can open its content.
- Every line, traced: Its instructions say to write only what a merged pull request or its changelog line shows, and to list the sources on the page.
- What each run cost: Tokens and cost are recorded per run and per agent, next to your plan's limits.
- A spend ceiling, on by default: Each run has a dollar ceiling unless you remove it. A run stops before it goes over it, and tells you.

### It can

- List and read releases and the pull requests in them
- Create draft Notion pages, under the page you name
- Post to one Slack channel you name, message you, or email you

### It can't

- Edit a release, comment, merge or push
- Publish the notes anywhere
- Edit, move or delete a Notion page
- Email anyone but you

## AI release notes, answered.

### What is a release notes generator?

A tool that writes release notes from what went into a release instead of from memory. Release writer reads the pull requests behind a GitHub release and writes notes for customers, not a list of pull request titles for developers.

### How is it different from GitHub's generated release notes?

GitHub can list a release's merged pull requests, its contributors and a link to the full changelog, grouped by labels you configure. Release writer reads each pull request and rewrites the change in plain words for customers, grouped New, Improved and Fixed.

### Does it publish the notes?

No. It saves a draft page in Notion, marked as a draft at the top, and posts a summary with the link. A person reads, edits and publishes the notes wherever your customers see them.

### Can it make up features?

Its instructions say every line must trace to a merged pull request or its changelog line, and the page lists the source pull requests at the bottom. A pull request it couldn't read is marked 'from changelog only' in the summary.

### What does it leave out?

Purely internal changes, such as refactors, CI changes and dependency bumps, unless a customer would notice them. The summary says how many it left out, so nothing disappears silently. Pull request numbers and author names stay out of the customer text.

### Which releases does it work with?

GitHub releases whose notes list their pull requests, such as the ones Release Please writes, tagged by component like web-v3.4.0. It finds the previous release with the same component name, so a repository with several components works.

### Does it change anything in GitHub?

No. It's given only GitHub tools that list and read releases and pull requests. It can't edit a release, comment, merge or push.

### Do I need to write code?

No. Describe the job in one sentence. Arclyx drafts the agent, you sign in to GitHub, Notion and Slack once, and you choose whether a webhook, a schedule or an @mention starts it.

## Agents that work well with it

- [Weekly eng update](https://arclyx.ai/agents/engineering-status-report): The week's work, risks and releases, for leadership
- [Incident brief](https://arclyx.ai/agents/incident-postmortem): What changed before it broke, in a postmortem draft
- [Blog drafter](https://arclyx.ai/agents/content-creation): A cited first draft of a blog post

[AI for engineering teams](https://arclyx.ai/solutions/engineering): AI for engineering teams that writes what surrounds shipping, from GitHub: the standup, the weekly report, release notes, postmortem drafts and issue triage.

## Early access

Build your release writer. One sentence in, a working agent out. Join the beta and we'll set you up. Join at https://arclyx.ai/#join or write to hello@arclyx.ai.

Last updated 24 October 2026.
