PI planning, short for planning-increment planning and often called big-room planning, is a recurring event where the teams that contribute to a shared outcome plan together for the next block of time. The value is the conversation: teams discover dependencies, negotiate scope, agree on commitments and leave with a plan they all understand. The cost is real. A large group in one room for a day or two is expensive, and most of the effort before and after the event is administrative.
That administrative work is where AI helps. It is not where the value lies, so shifting it to agents frees time for the part that matters. This guide describes how to do that without turning the event into a document-generation exercise.
What the event is for
- Sharing the business context and the priorities for the next period.
- Letting teams draft plans and surface risks and dependencies.
- Negotiating across teams until the plan is feasible.
- Making commitments in public, with confidence levels.
- Agreeing on how the plan will be tracked and adjusted.
Before the event: agents as preparers
- Gather the inputs. An agent can assemble current priorities, the backlog state, last period's outcomes and open risks into a single briefing, with sources linked, for teams to read in advance.
- Draft candidate scope. Based on the backlog and capacity, an assistant can propose a first cut of what each team might take on, clearly labeled as a draft for the team to rewrite.
- Pre-check dependencies. An agent can scan items across teams and flag likely dependencies so teams start the event with a list to confirm or dismiss.
- Prepare the room. Logistics, schedules and reading material can be handled by agents so organizers are free to coach.
During the event: keep it human
We recommend using AI sparingly during the event. The room is for talking. A few uses help, though.
- A note-taking agent captures decisions, risks and commitments in a consistent structure, which people confirm before leaving.
- A dependency agent keeps a live list of cross-team asks that have not yet been answered, so nothing is lost between breakouts.
- A consistency checker highlights plan items with no outcome attached, or commitments that exceed stated capacity, as questions to discuss.
- An agent can answer factual questions from the briefing, such as “what did we decide about this last time?”, so people do not have to search.
What the agents should not do is decide priorities, make commitments or settle disputes between teams. Those are negotiations among people who must live with the result.
After the event: follow-up that actually happens
- Publish the plan. An agent converts the confirmed commitments into the tracking tool and links them to objectives, so the plan and the work are the same record.
- Watch the plan. Agents can flag items that are slipping or dependencies that have not moved, and prompt the owners to talk early.
- Summarize for leaders. A short, plain-language status drawn from the real data replaces a slide deck that nobody trusts.
- Capture the learning. At the end of the period, an agent drafts a review of what was planned versus what happened, which the teams edit and keep.
Connect planning to outcomes
A plan that lists features is easy to produce and hard to judge. Tie each team's plan to the outcomes it is meant to move, using objectives and key results. Then the planning conversation is about what change you expect and how you will know, not only what will be built. Our book Outcome-Driven AI develops this idea further.
Design the human-and-agent roles
Name the people responsible for each agent used in planning: who maintains its instructions, who checks its output, who decides what it may see. Treat these like any other role at the event. Patterns for dividing work between people and agents are collected in 100 AI-Agile Teams.
Pitfalls
- Using agents to produce more paper, not better conversation.
- Presenting an agent's draft as the plan before teams have owned it.
- Hiding dependencies in an automated report rather than discussing them.
- Skipping the review of the previous period because the agents make the next plan so easy.
If your teams are not yet clear about where value flows across the organization, begin with a value stream identification workshop. To tackle process problems that planning reveals, see Kaizen events for AI-enabled work.