Release Notes From Sprint Demo Calls, With the Team's Own Words

The release goes out on Tuesday and the notes are still a blank page. What went into it was said out loud: two sprint demos, a developer showing each change, someone mentioning on the way that recurring jobs still do not sync in one case. The commit log has the code, not the explanation, and the tracker says "Done" in the developers' shorthand. The best description of each change is what the developer said while showing it.

A shorter route: record sprint demos in Speak-Y, press a hotkey when a developer shows something that ships or names a known issue, and before the release ask your AI assistant to collect the marks over MCP for the release tag and draft the notes. Each line comes back with who said it, in their words, the demo date and the time in the recording, and the ticket to check. Confirming what ships, writing for customers and promising dates stay with people.

This guide follows the Speak-Y macOS app and its MCP server as checked on 17 November 2026, with one invented example: Priya Nair, product manager at Crewboard, a scheduling app for field service teams, ships release 4.2 to customers on 24 November. It contains the work of two sprints, shown at demos on 30 October and 13 November.

What do release notes need from a sprint demo?

Release notes guides agree on the shape. Keep a Changelog (version 1.1.0) starts from "changelogs are for humans, not machines", groups changes by type — Added, Changed, Deprecated, Removed, Fixed, Security — and warns against dumping the git log into them. MongoDB's documentation style guide keeps a Known issues section in every set of release notes, even when it says "None for this release", and asks for each issue's effect on the user and any workaround (both searched on 17 November 2026).

The demo is where these facts are said out loud. The Scrum Guide describes the Sprint Review as a working session with stakeholders, not only a presentation — so the explanation, the questions and the caveats are all on the call. Four things get lost between the demo and the notes:

Before the first demo: two note types and a release tag

Open Settings → Notes. Under Note types, rename the default Mark the moment to Shipped and keep it on ⌥1. Click Add note type, type Known issue into Type name, click Save, then Set hotkey and press ⌥2.

Pick one tag for the release, such as release-4-2, and put it on every demo whose work goes into it. Before the release the assistant then reads exactly these calls, and nothing from an earlier release or a customer call.

What each mark becomes in the release notes: Shipped and Known issue marks, and the hedged "should be in" that is not a mark
Two note types sort the demo while it happens; a maybe waits for the tracker.

During the demo: let the developer show it, press the key

Record the demo as a meeting. No bot joins: Speak-Y records your microphone and the call's sound on your Mac, so it works the same in Zoom, Google Meet or Microsoft Teams. If people outside the team attend, say at the start that the demo is recorded — see is it legal to record meetings for how consent rules differ.

Three habits make the marks useful:

  1. Press after the sentence. A mark covers the last 15 seconds of the call, so the description just said lands inside it.
  2. Ask "what can the user do now?" It turns a demo of an endpoint into a sentence a customer understands. Mark the answer on ⌥1.
  3. Ask about the edges. "Does it work for recurring jobs?" brings out the caveat while the developer is still on screen. Mark it on ⌥2.

At the 13 November demo, Priya marks two changes and one known issue. At 08:15 the mobile developer Hana Kim says, "Technicians can now reorder their day by dragging jobs in the mobile app, and the dispatcher sees the new order right away." Asked at 19:35 whether it works for recurring jobs, Hana answers, "Recurring jobs still don't sync to Google Calendar if the series has an exception. You have to edit the event by hand." Priya marks it as a known issue. When the backend developer Marco Lee adds that CSV export "should be in, if QA passes it on Friday", she does not mark it: it is something to check, not something to announce.

After the demo: name the speakers, add the tag

On By Speakers, click each speaker chip and type the name and role — "Hana Kim, mobile" — so the draft quotes people, not numbers. Add the release-4-2 tag. The chips above the transcript count the marks by type, and a click on a chip shows only those replies.

The Speak-Y recording of the 13 November sprint demo on By Speakers, with Shipped and Known issue marks next to the developers' replies
Each mark sits next to the reply it belongs to, with the developer's name and the time in the demo.

The same recording serves the next meeting too: the retrospective can start from it, as in sprint retro follow-through.

Ask your assistant for the draft

Connect your assistant to the Speak-Y MCP server once: Connect AI assistant in the account menu at the bottom of the sidebar, or Settings → Integrations, then Connect next to Claude, ChatGPT or Cursor. The steps are in how to connect meeting notes to AI assistants. The server is free on every plan and reads the library on your Mac.

Save standing instructions in a project in Claude or ChatGPT, so before each release the request is one line:

Draft the release notes for Crewboard 4.2:
demo calls tagged "release-4-2".

Rules:
- Use only my Shipped and Known issue marks.
- Group as Added, Changed, Fixed, Known issues.
- For each: the developer's words, who said it,
  the demo date and the time, like (Nov 13, 08:15).
- Add "Check:" with the ticket to confirm. Never
  invent a ticket number.
- List anything said with "should be" or "I think"
  as unconfirmed.
- No customer-facing wording and no dates.

The assistant calls list_recordings with the tag and reads each demo. Alongside the transcript, Speak-Y hands it a Notes list — each mark with its time, type, speaker and first words. A mark's time is the moment you pressed the key, a few seconds after the reply began; the draft cites the start of the reply:

Release 4.2 — draft (from sprint demos)
Added
- Drag to reorder the day. Hana Kim: "Technicians
  can now reorder their day by dragging jobs in the
  mobile app." (Nov 13, 08:15)
  Check: reorder ticket, mobile release
Changed
- Job reminders. Marco Lee: "Reminders now go out
  an hour before the job, not the night before."
  (Oct 30, 14:05)
  Check: reminder settings ticket

Known issues
- Hana Kim: "Recurring jobs still don't sync to
  Google Calendar if the series has an exception.
  You have to edit the event by hand." (Nov 13, 19:40)

Unconfirmed
- CSV export. Marco Lee: "Should be in, if QA
  passes it on Friday." (Nov 13, 23:10)
What the assistant reads over MCP — the marks with times, types and speakers — and the release notes draft it groups them into
Changes are grouped by type; every line keeps the demo and the time it was said.

The demos do not know what happened after them. If QA moves CSV export to the next release, say so in the same chat: hold Right ⌘ in the message field, dictate "CSV export didn't pass QA, it moves to 4.3", release and send. The assistant takes the line out of the draft and keeps what was said at the demo next to it.

What stays with the product manager

The draft knows what the team said. Four things it cannot do:

Four checks before the draft becomes published release notes, each with the red flag that shows it is not ready
The demos tell you what was said; the tracker, the wording and the promises belong to people.

Limits and costs

The short version

Shipped on ⌥1, Known issue on ⌥2, one tag per release. Ask what the user can do now, ask about the edges, press the key; then speaker names and the tag. Before the release one request gives a draft grouped by type, with the developers' words, times and the ticket to check; what ships, the customer wording and any promise are yours. To turn the planning meeting at the other end of the sprint into tasks, see sprint planning to task drafts in Claude Code.

FAQ

Can an AI assistant draft release notes from recorded sprint demos?

It can draft them from what the team said. Connected to the Speak-Y MCP server, Claude, ChatGPT or Cursor reads the demo calls with your release tag — the transcript by speaker and a Notes list of your marks with their times — and groups the marked moments into Shipped and Known issues, with the developer's words, the demo date and the time in the recording. What actually ships is then checked against the tracker.

How do I mark what shipped during a sprint demo?

Create note types in Speak-Y under Settings → Notes: rename the default Mark the moment to Shipped on ⌥1 and add Known issue on ⌥2. A press marks the last 15 seconds of the call with that type, so the sentence the developer just said lands in the assistant's list sorted, with the speaker and the time.

Is a demo enough to say a feature is in the release?

No. A demo shows that something works on someone's machine; the tracker and the release branch say whether it ships. Ask the assistant to add a Check line with the ticket to confirm for every item, to list anything said with 'should be' or 'I think' as unconfirmed, and never to invent a ticket number.

Should release notes keep the developers' wording?

Keep it in the draft, not in the published notes. The developer's sentence is the most accurate description of the change and the time in the recording lets you replay it; the customer-facing wording is rewritten by the product manager or writer, in terms of what the user can now do.

Which Speak-Y plan do I need for 45-minute sprint demos?

Pro or higher. Checked on speak-y.com on 17 November 2026: Free stops a meeting recording at 30 minutes, Pro ($10 a month) at one hour, and Pro+ ($15) and Max ($30) at four hours. The MCP server is free on every plan.