We only use cookies to enhance your experience. No cross-site tracking, no ads. Read our Cookie Policy.
One prompt fetches every comment on your recording and gives you the summary and the actions in one go.
Record a supercut showing your work and ask for comments on the timeline.
Your agent fetches every comment, reaction and timestamp.
Get the actionable insights, review them all and push the next steps.
The fastest way to collect video feedback from stakeholders is to record a supercut showing your work, ask your stakeholders to add timestamped comments on it, and then let your agent fetch the comments and write up actionable insights you can review and push to Linear or Notion.
Connected to your agent.
Connected to your agent. Either one, or both. The summary fits a Notion page, the actions fit Linear issues.
Claude, Cursor, or similar.
Share the recording and tell people to comment on the moment rather than at the end. A comment left at 1:42 is attached to what was on screen at 1:42, and that is what makes it fetchable later as something specific rather than a general opinion. Three things worth asking for:
Not a reply in a thread somewhere else. Feedback that lives with the recording is feedback your agent can read.
Ask people to say why. "This bit is confusing" turns into an action. "Hmm" does not.
People who will not write a paragraph will still react at the part they liked, and that gets fetched too.
Give your agent the supercut link. It pulls the comments with their timestamps and authors, the reactions, and the transcript, then lines up what each person said against what you were showing at that second. What comes back is two lists: the summary of what people said, and a numbered set of actions. Nothing gets written anywhere yet. It groups the feedback by theme rather than by person, which is what turns twelve comments into three things to decide.
Read the actions and pick the ones that are real. Some will be the same point twice, some will be someone thinking out loud. You decide what becomes work. Then one more prompt sends the ones you approved where they belong, and every point still links back to the second it came from so nobody has to trust the paraphrase.
One issue per action, with the supercut attached and the comment that asked for it. Your agent reads your real labels, statuses, teams and cycles first, so the issue lands with your values rather than invented ones.
One page with the summary at the top and the approved actions underneath as a checklist. Better when the round of feedback itself is the thing worth keeping.
The first prompt tells it to keep both sides. A summary that quietly resolves an argument nobody resolved is worse than no summary.
Two prompts, because you read the list before anything gets created. The first one fetches and summarizes. The second one pushes only what you approved.
Share your recording, give people time to comment, then paste this. Swap [SUPERCUT LINK] for your recording.
Here is a supercut I shared for feedback: [SUPERCUT LINK]
Fetch everything people left on it and show me the summary and the actions. Do not write anything anywhere yet.
1. Fetch the comments on that recording, with the timestamp each one was left at and who left it.
2. Fetch the reactions. A reaction with no comment still tells you which moments landed.
3. Fetch the transcript, so you know what was on screen and what I said at the second each comment was left.
4. Group the feedback by theme, not by person. If three people said the same thing in different words, that is one point with three names on it.
5. Show me two lists.
The summary:
- What people agreed on
- What people disagreed on, with both sides
- Questions nobody answered
- A link to the second in the recording for every point
The actions, numbered so I can pick from them:
- One line each, written as something a person can actually do
- Who asked for it
- The link to the second it came from
Rules:
- Quote people in their own words where the wording matters. Do not smooth it into consensus that is not there.
- Keep disagreement visible. A summary that hides it is worse than no summary.
- Do not invent actions. Only list one if someone actually asked for something.
- If a comment is ambiguous, list it as needing clarification rather than guessing what it meant.
- Stop after the two lists and wait. I will tell you which actions to push.Reply with this once you have picked. Swap [NUMBERS] for the actions you want and [LINEAR OR NOTION] for where they go.
Push actions [NUMBERS] to [LINEAR OR NOTION].
If it is Linear, create one issue per action:
- Title: the action, in the words the person used
- Description: what they said, who asked for it, and the link to the second in the recording
- Attach the supercut, so whoever picks it up can watch the moment
- Read my labels, statuses, teams and cycles first and use those. Do not invent values.
If it is Notion, write one page:
- The summary at the top, grouped by theme
- The actions I approved underneath, as a checklist
- Every point linking back to the second it came from
Rules:
- Only push the actions I named. Leave the rest in the list.
- Do not reword an action into something softer or bigger than what was asked for.
- If an action is already an open issue, add a comment with the new feedback instead of filing a duplicate.
- When you are done, tell me what you created and where.The agent can only work with what people left. Ask for comments on the timeline and the first prompt has something to read.
Asking for feedback is easy and acting on it is easy. The part that breaks the loop is the bit between them: reading twelve comments left over three days, working out which ones are the same point, and deciding what any of it means. That is the step people put off, and a feedback loop that stalls there is the same as no feedback loop at all. Fetching it and reading it is now one prompt. Ask, wait, run it, and the summary and the actions are in front of you. You still decide what happens next, which is the part worth keeping.
A product team reviewing a prototype, a sales team reviewing a pitch, a support team reviewing a help article. Anywhere people look at a screen and react to it.
Gathering feedback by hand gets harder with every person you add. Fetching it does not, so you can ask more people without paying for it later.
This is one of several things your agent can do across the two tools.
Record the bug happening. Your agent gets the transcript and frames, and the Linear ticket writes itself.
Explain how it should work, once. Your agent writes the stories your developers need to build it.
Point your agent to your PM's feature demo video, and it writes the help center article, screenshots included.
Record yourself doing the work once. Your agent picks it up and writes the documentation directly in your database.
Record your update. Your agent posts the summary, the action items, and links to the moments that matter.