AI Estimation: Planning Poker for AI Teams

Story points tell you how much work a story is. A team that builds with AI also wants to know how many tokens it will burn and whether an agent can build it at all. Storyvote asks all of it in one free planning poker round: everyone picks a points card and answers the AI questions, and every answer stays hidden until the reveal.

Four questions next to the estimate

A room can ask any of these beside its main deck. Each has its own short row of cards, and the answers are optional, so a vote counts as soon as the points card is picked.

Not sure what a 250k story looks like next to a 2.5M one? Our AI token estimation guide goes through the scale card by card.

How a round works

1. Create a room.Start one with your points deck, and turn on AI tokens and Agent-ready under extra questions (or add them later in Settings). No sign-up: share the link and people join with a name.
2. Vote on your own. Each person picks a points card and answers the AI questions. Votes are held on the server until the reveal, so nobody anchors on someone else's number, or on the one an AI assistant suggested.
3. Reveal and talk about the spread. Every question shows how the votes fell. A story voted both 100k and 5M, or both Agent and Human, is the one worth a conversation.
4. Keep the result. The most-picked card of each question is saved on the story (on a tie, the more worrying one), and the CSV download has a column for each, ready for the sprint's token budget.

Our guide to planning poker for AI teams walks through the whole setup, and how to read a reveal with two numbers.

Why estimate tokens as a team

• The cost of a story moved. Two 3-point stories can differ fifty times in tokens: one needs a file of context, the other sends an agent through half the repository. A quick forecast tells a cheap ticket from an expensive one before anyone prompts.
• Who builds it is a planning decision. A sprint of Agent stories and a sprint of Human stories need different people and different review time. Agent-ready makes the split visible while you plan.
• The disagreement is the signal. When one person votes 50k and another 5M, they are imagining different work. Finding that out costs a minute in the meeting, and a lot more after an agent has run with the wrong idea.
• An AI can't vote for your team. An assistant can draft stories and suggest sizes, but it doesn't know your codebase's sharp edges or who is on holiday. Here's what to hand to the assistant and what to keep.

Estimate in tokens only

Each of these scales is also a voting system of its own. Pick AI tokens as the room's deck to vote on token usage alone, for a team that budgets agent work separately from the people's. Everything else works the same: hidden votes, the reveal, the discussion timer, live or async rooms, and stories split by discipline.

Guides for AI teams

• Planning poker for AI teams: how to set up the room. A practical setup for teams that build with AI tools: story points plus AI tokens, Agent-ready, risk and clarity, and how to read a reveal with two numbers.
• AI token estimation: a card-by-card guide for agile teams. How to estimate AI token usage with your team: what each card from 10k to 10M+ usually means, reference tickets, and what to do when the votes split.
• Why a million AI tokens never has one price. Input, cached input and output tokens are priced differently, so one token estimate can cost three times as much. A worked example and a sprint worksheet.
• Budget for the retry: estimating AI agent tokens before you prompt. Why an AI agent's retries cost more each time, how to estimate a story's token budget with planning poker, and tripwires that stop a runaway task.
• Planning poker when AI agents write the code. AI coding agents build fast but never ask questions. Why estimating together matters more now, and how to turn a planning poker round into an agent-ready spec.
• Sprint estimation with Cursor and other AI coding tools. What AI coding tools change in a sprint and what they don't: estimating the work, not the tool, context-heavy stories, review capacity and velocity.
• AI sprint planning: what to hand to the assistant and what to keep. Where an AI assistant helps sprint planning, what it cannot know about your team, and an order of play that keeps the vote independent of its number.

Questions

How do you estimate AI tokens in planning poker?

Turn on the AI tokens question when you create the room, or later in Settings. Next to their story points, each person picks a card from 10k to 10M+ for how many tokens an AI agent would spend on the story, from the first prompt to the finished pull request. The cards stay hidden until the reveal, and the most-picked one is saved on the story.

Does Storyvote measure our real token usage?

No. It records the team's forecast. It doesn't connect to your AI tools or read your bills, so compare the estimates with your provider's usage after the sprint.

What does Agent-ready mean?

It asks who should build the story: Agent (an AI agent can do it on its own), Pair (a person working with an agent) or Human (a person, without one). A "?" card means the voter can't tell yet.

Can we estimate in tokens only, without story points?

Yes. The AI tokens scale is also one of the voting systems, so a room can use it as its deck and vote on tokens alone. Most teams keep points for the human work and add tokens as a second number.

Is it free?

Yes. Every feature, the AI questions included, is free for every team, with no sign-up, no credit card and no player limit.