Planning poker for AI teams: how to set up the room

15 October 2026 · Updated 19 October 2026 · 8 min read

Planning poker still works when half the code is written by AI tools. The team still needs to agree on what a story is and how hard it will be, and that conversation matters even more when an agent will take the ticket literally. What changes is what you ask in the round. This guide is the practical part: how to set up a room for a team that builds with AI, which questions to add, and how to read the reveal when every story comes back with more than one number.

Who this is for

Engineering teams, tech leads, Scrum masters and product owners whose stories regularly go through Cursor, Claude Code, Copilot or a similar tool. If AI usage is now a cost you watch, or it decides whether a person or an agent picks up a ticket, the setup below is worth the five minutes. If it isn't, plain story points are still the right tool.

Pick the questions

Keep story points as the main estimate: they measure effort, complexity and risk for the team, and review time is still mostly human time. Then add the questions that fit how you work. In Storyvote they're called extra questions, and each has its own small deck:

Start with one or two. Four questions on every story slows a round down; most teams settle on points plus tokens, or points plus Agent-ready. The extra questions are optional for each voter, so someone who has no feel for token usage yet can still vote on points.

Set up the room

1. Create a room on the new room page and choose Live for a planning call, or Async if the team votes across time zones over a day or two.
2. Choose the voting system: Fibonacci, modified Fibonacci, T-shirt sizes, powers of two, or your own cards.
3. Add your extra questions from the list. You can change them later in Settings, between stories.
4. Optional: split stories by discipline (say FE, BE and QA) if each side estimates its own part, and start the discussion timer to keep rounds short.
5. Bring in the stories: add them one by one, import a sprint from Jira or a Trello list, or import a CSV from any other tracker or spreadsheet.
6. Share the link in Slack, Teams or wherever the team talks. Nobody needs an account; Storyvote also runs inside Telegram and Discord.

One round, every number

Each person picks a points card and, if they want, a card for each extra question. Everything stays hidden until the reveal, then turns over at once. Revealing together matters twice over here: token estimates are new to most teams, and the first number said out loud tends to become everyone's.

After the reveal, each extra question keeps its most-picked card as the story's answer (the more worrying one on a tie), shown on the story in the list next to the estimate. When you download the stories as CSV, each question gets its own column.

Reading two numbers together

The interesting stories are the ones where the numbers disagree with each other. A quick guide to points against tokens:

Agent-ready adds a third angle. A story the team calls Agent but votes Vague on clarity isn't ready for an agent at all; it's ready for a conversation. A Human story with a low token card is fine: the person might still use AI for parts of it.

When the votes split

A spread on points is about the work. A spread on the extra questions usually points somewhere more specific:

• Tokens split: someone knows about context the ticket doesn't mention. Write it in. Our guide to AI token estimation walks through the scale card by card.
• Agent-ready split: people pictured different approaches. Agree on the approach, then vote again.
• Clarity split, or a few ?: the spec has a hole. Refine the story instead of averaging the guesses.

The same habits that help with any reveal apply here; see how to read a reveal for the full version.

From planning to the backlog

Over a few sprints, the extra answers become useful data. The token cards add up to a rough forecast of AI spend. The Agent-ready answers show how much of the backlog you actually hand to agents, and which areas never qualify. And setting the cards next to real usage at the retro tells you where the estimates, the tickets or the agent instructions need work.

If you're coming from another tool, our checklist for switching covers moving your backlog across.

Try it with your team

Create a room, keep your points deck, add AI tokens or Agent-ready, and run your next refinement with it. It's free, with no limits on players or stories, and the room is deleted when you end the session. The AI estimation page sums up every question a room can ask.

← All articles