Sprint planning with AI agents is now a practical question, not a thought experiment. Jira, Linear and GitHub all let teams assign tickets to coding agents, and “Scrum is dead” has become a favourite take on LinkedIn. The more useful question for a Scrum Master is narrower: when some backlog items will be worked by an agent, what changes in the planning meeting, and what stays the same?

This article extends our sprint planning guide. For the non-planning half of the role, see AI Scrum Master: what it can run for you.

What stays the same

Three things in the Scrum Guide do not move because an agent joins the board:

• The sprint goal. The Scrum Team decides why the sprint is valuable. An assistant can draft a goal from the selected items, but a person has to accept it. See how to write a sprint goal.
• Selection of work. Developers select items and plan how to do them. An agent can propose; the team decides.
• Definition of Done. Humans stay accountable for review, tests and release. Agent output that has not been reviewed is not Done.

What changes: the constraint moves to review

When an agent drafts a pull request in minutes, writing code is no longer the slowest step. The slow steps become specifying the ticket well enough for the agent, reviewing what it produced, and fixing what it got wrong. All three are human time. A team that plans a sprint as if agent-assigned tickets cost nothing will overcommit its reviewers, and the work will pile up in “In Review”.

The planning rule that follows: every agent-assigned item still consumes human capacity. Estimate that capacity explicitly.

How to estimate agent-assigned work

Giving an AI agent story points and adding them to velocity mixes two different kinds of throughput. A cleaner method:

1. Split the item. For each agent-suitable ticket, create or estimate the human parts: “specify and prompt”, “review and test”, “integrate”. Estimate those in hours or points like any other work.
2. Assign the human parts to named people. Review time lands on specific reviewers, usually the most senior developers, who are also the scarcest.
3. Keep agent throughput separate. Track how many agent-assigned items reached Done per sprint as its own number. Do not blend it into team velocity until you have several sprints of data.
4. Hold a rework buffer. Early on, some agent output will be rejected. Leave slack for it, the same way you would for unplanned work.

Backlog refinement matters more, not less

An agent works from what the ticket says. Vague acceptance criteria that a colleague would clarify over chat produce wrong output from an agent. Teams that get value from agents tend to refine more carefully: small, well-bounded items, explicit acceptance criteria, and clear notes on what the agent must not touch. Refinement is where you decide which items are agent-suitable at all.

A planning meeting checklist for mixed teams

• Which items are agent-suitable, and who decided?
• Who reviews each agent-assigned item, and is that time in their capacity?
• Is any reviewer above 100% once review time is added?
• Does the sprint goal still hold if agent output on the key item is rejected?
• How will agent throughput be reported separately at the review?

These questions slot into the capacity step of the sprint planning meeting agenda.

Doing it in Jira

The risk above is a capacity problem: review and specification hours landing on a few people. Sprint Planning, Capacity & Resource Planning for Jira by Divim shows each team member’s assigned estimates against their available time for the sprint, including days off, so a reviewer who is over-allocated by agent reviews is visible before the sprint starts. Subtasks such as “review agent PR” can be created, estimated and assigned from the same planning screen.

If you are setting this up for the first time, our guide to AI-assisted sprint planning in Jira walks through the full planning session step by step, including where AI helps and where the team still has to decide.

For the refinement side, Backlog Refinement for Jira helps the team size and prepare items before planning, which is where agent-suitable work gets identified.

Summary

• Sprint goal, selection and Definition of Done stay human.
• Agent-assigned items still cost human time: specify, review, integrate.
• Estimate the human parts and assign them to named reviewers.
• Track agent throughput separately from velocity.
• Refine more carefully; the ticket is the agent’s only context.

Back to the sprint planning guide.