Release notes that write themselves
from the commits you already made
Release notes get written in a rush right before a ship date, or skipped entirely, because nobody wants to scroll through two weeks of commits to remember what actually changed. We build an agent that reads every merged pull request and drafts the changelog in your product's voice, ready for a person to approve.
The process today
Release notes are usually the last thing written before a ship date and the first thing cut when time runs short. A developer who wrote a fix two weeks ago has to reconstruct what it was for from a commit message that reads “fix bug,” and a product manager who wants customer-facing copy ends up asking three people in Slack what actually changed. Teams that track this honestly often find release notes take an hour or more per cycle to assemble, not because the writing is hard, but because the information is scattered across pull requests, tickets and Slack threads that nobody wants to re-read.
The cost shows up downstream. Customers who do not know a bug was fixed keep filing the same ticket. Support teams repeat answers that a changelog would have already covered. Internal teams miss a breaking change buried in a bullet list that looked the same as every other line. The common fix, assigning release notes to whoever has time that week, produces notes with no consistent tone and no reliable record of what shipped when.
What the agent does
Reads every merged pull request and linked ticket since the last release, pulling the actual diff and description rather than relying on a commit message alone, so a one-line commit still gets a readable explanation.
Drafts two versions in one pass, a customer-facing changelog in plain language and an internal version with the operational detail (migration steps, flags, rollback notes) that engineering needs and customers do not.
Groups changes by type, features, fixes and breaking changes kept visually distinct so a reader scanning the list finds what matters to them in seconds, with breaking changes flagged separately for extra review.
Matches your product’s existing voice, learned from past release notes you already publish, instead of a generic “we made some improvements” tone that says nothing.
Links every entry back to its source, the pull request or ticket it was drafted from, so a reviewer checking a claim does not have to go hunting through git history.
Schedules the draft ahead of the cut, timed to your release cadence (weekly, per-sprint, per-tag) so the draft is sitting ready for review instead of being written under pressure on ship day.
What stays with humans
The agent drafts, a person publishes. Tone calls on sensitive changes (a price change, a feature removal, a security fix that needs careful wording) stay with whoever owns customer communication. Breaking changes are always flagged for a second look before anything goes out, and nothing reaches a customer-facing channel without a human clicking publish.
Guards
A review period where drafts land in a shared doc or a Slack channel instead of the live changelog, so your team checks the agent’s tone and accuracy against a few real releases first. Every draft logs which pull requests and tickets fed it, so a factual claim can always be traced back. Rate limits keep the agent from generating a new draft more often than your release cadence calls for, and a kill switch pulls it off any repo instantly.
Price and timeline
| Package | Price | Best for |
|---|---|---|
| Single automation | from $500 | One repo, drafting customer-facing and internal release notes from your pull request history |
| Department package | from $2,500 | Release notes plus code review, test generation and log triage for the same codebase |
3 to 7 days, most of it spent learning your existing release notes so the first draft already sounds like your product instead of a generic bot.
Related
Pairs with code review and test generation to close the loop from code change to customer-facing copy, and with log triage and alerting so a fix that shows up in the logs also shows up in the next changelog. Part of automation of everything digital and built the way we build AI agents for our own products. We run this discipline on our own releases, including the visa slot monitoring bot with OCR and the Telegram game club with 14 engines.
Tell us which repo and release cadence you want covered and we will send back a fixed price and a plan for the first week: get in touch.
Tired of doing this by hand? We can take the whole routine off your team, not just this step: Routine takeover, from $400 →
FAQ
How much does an AI release notes agent cost?
A single-repo agent drafting changelogs from your existing pull request history starts from $500. A department package covering release notes alongside code review, test generation and log triage starts from $2,500.
How long does it take to go live?
3 to 7 days: connecting to your git provider and ticket tracker, learning your product's existing tone from past release notes, and a short review period before the agent's drafts go out without a full rewrite.
Which tools does it connect to?
GitHub, GitLab and Bitbucket for commit and pull-request history, Jira, Linear or your own ticket tracker for context, your existing changelog platform or CMS, and Claude for turning a list of commits into readable prose.
What if the agent describes a change wrong?
It drafts, it does not publish. Every entry links back to the pull request or ticket it was drawn from, so a reviewer can check the claim in one click before anything goes out to customers.
Is our commit history and roadmap safe?
The agent reads only the repositories and trackers you connect it to, under your own provider's permissions, and we do not retain your commit history or ticket content outside the drafting session.