AI sprint planning: what to hand to the assistant and what to keep

13 October 2026 · 5 min read

An AI assistant can read a backlog faster than any of us, and it will happily give you a number for every story in it. The risk is not that the number is wrong. It is that the first number in the room quietly becomes everyone's number. Here is where an assistant earns its place in sprint planning, where it doesn't, and an order of play that keeps the team's own judgement intact.

What an AI assistant does well

• Tidying the backlog before the meeting. Turning a one-line ticket into a draft with a goal and acceptance criteria, merging duplicates, and splitting an obviously huge story into pieces for the team to react to.
• Spotting gaps. "What happens if the user has no saved card?" is the kind of question an assistant asks tirelessly, and it is cheap to ask before the story is estimated rather than after.
• Reading the room's history. Pasted in, last sprint's stories and what they really took make a decent prompt for "which of these look like the ones that slipped?".
• Summarising. A long discussion becomes three decisions and two open questions, written down before everyone forgets which was which.

What it can't do for you

• Know your team. It doesn't know that the payments module is half-migrated, that your best reviewer is away, or that nobody has touched the legacy export in two years.
• Know your capacity. Holidays, support rotation and the person who is learning the codebase are in your head, not in the ticket. See capacity planning for Scrum teams.
• Own the commitment. A sprint is a promise the team makes to itself. A number the team didn't argue about is a number nobody feels bound by.
• Replace the disagreement. The useful part of an estimate is the gap between the 3 and the 13. An assistant's single answer erases the gap before anyone has seen it.

The order of play that works

The same assistant helps or hurts depending on when it speaks. Put it before the vote to clear the ground, and keep it out of the vote itself:

1. Prepare with it, before the meeting. Use the assistant to sharpen the tickets, split the oversized ones and list the open questions. Treat the output as a draft for a person to edit.
2. Read the story out, then vote in secret. Everyone picks a card independently, and nobody, human or machine, shares a number first.
3. Reveal together and look at the spread. Agreement moves on. A wide spread is the signal to talk. Here is how to read a reveal.
4. Only now, ask the assistant. Give it the story and the cards that were played, and ask what the people on the low end and the high end might each be seeing. It is a cheap way to surface the assumption behind a 13.
5. Decide as a team, and write down why. Revote if the discussion changed minds, then record the estimate and the one-line reason, so the retro has something to check it against.

A few habits that keep it honest

• Never show an AI number before the cards are down. It anchors, however sceptical you feel about it.
• Keep a count. Note when the assistant was right and when the team was. After a few sprints you will know which kinds of stories it reads well.
• Estimate the work, not the tool. If the story will be built with an AI coding tool, that changes how it is reviewed and how it can go wrong, not what the team is being asked to size. We cover that in sprint estimation with AI coding tools and planning poker when AI agents write the code.
• Mind what you paste. Story text and names can be confidential. Use only what your organisation allows you to share with an assistant.

Where Storyvote fits

Storyvote has no AI that votes or suggests a number, on purpose. What it gives you is the part of the order of play that tools usually get wrong: votes held on the server until the reveal, so nobody can peek or be nudged, and the spread shown clearly once the cards turn over. Use whichever assistant you like beside it, before the vote for the ground work and after the reveal for the questions. For the rest of the meeting, see our sprint planning guide.

← All articles