Adding AI cost to planning sounds like it needs a new process. It does not. An AI-aware planning session is your normal planning poker, plus one extra vote per ticket. Here is exactly how to run your first one, from setup to a shared sprint plan, in about the same time your current session takes.
Before the session: agree on a token deck
The only prep that matters is making sure everyone votes against the same token tiers. Pick a deck and anchor each tier to a recently shipped ticket. If you do not have one yet, start from the token estimation tiers and adjust the names to fit your team. Five minutes here saves a lot of confusion mid-session.
1. Create the room and bring the backlog
Spin up a room and share the link — no account needed, so everyone can join in seconds. Drop in the tickets you want to estimate. Keep the list short for the first session; ten to fifteen tickets is plenty to build the habit without exhausting the team.
2. Vote on effort, then on tokens
For each ticket, run the round you already know: everyone privately votes story points, then reveal together. Then do a second, quick vote on the token tier using your agreed deck. Keeping the two votes separate is deliberate — it preserves the familiar effort discussion while making AI cost explicit instead of a hidden assumption.
3. Mine the disagreements
The reveal is where the value is. Watch for two patterns:
- Split effort votes — the usual signal of hidden complexity or differing scope assumptions. Talk it out before locking.
- Split token votes — people are assuming different context, models, or implementation paths. This conversation is often new for teams and surfaces real cost risk early.
- Effort and tokens that diverge — a cheap-looking ticket with a high token tier is your most important flag. Do not let it slip through as an easy win.
4. Turn the estimates into a plan
Once every ticket has both numbers, use them together. Re-prioritize low-effort, high-token tickets, pick a cheaper model or tighter context where quality allows, and roll the token tiers up into a rough AI budget for the sprint. The full decision framework is in how to estimate AI coding costs during sprint planning.
5. Keep it lightweight and repeat
Do not add ceremony. The token vote is one extra signal in a round you already run, and it gets faster every session as your anchors settle. If you are still deciding whether the second estimate earns its place, read why AI teams may need both story points and token estimates. After a sprint or two, cost awareness simply becomes part of how your team plans.