Human/AI attribution
Scrum Manager's guide Scrum in Teams with AI describes this practice as the most important adaptation of Sprint Planning in hybrid teams. Without an explicit attribution, the team tends to fill the sprint with AI generation tasks without reserving enough capacity for human review, which collapses the flow at the end of the cycle.
The three estimation levels
Human/AI attribution classifies each backlog item into one of three levels:
| Level | Type | Description | Who does it |
|---|---|---|---|
| Level 1 | Automated | Task the agent can carry out autonomously with minimal supervision. The spec is complete and the agent has everything it needs. | AI agent (with HITL/HOTL supervision according to risk) |
| Level 2 | Human | Task that requires human judgement, organizational context, creativity, or strategic decision-making that AI cannot provide. | Team member |
| Level 3 | Review and integration (R&I) | Human work on AI output: verifying, integrating, validating, and approving what the agent has generated. | R&I task → team member |
How it works in sprint planning
The human/AI attribution flow within Sprint Planning follows these steps:
- The product architect presents the prioritized backlog items for the sprint.
- The team classifies each task according to the three levels.
- For level 1 tasks (automated), the Definition of Ready for AI is checked: does the agent have everything it needs? Is the spec complete? Are the boundaries defined?
- The specs required for level 1 tasks are written or refined.
- The sprint is planned explicitly including the R&I tasks (level 3), not only the generation ones.
- The Sprint Goal is set with a strategic focus.
The R&I task rule
The most common mistake in sprints with AI is planning them as if the work ended with generation. Human review of AI output is not an "extra" to be done if there is time left over: it is a first-class task that consumes real team capacity and must be planned explicitly.
The Scrum Manager guide sets a reference ratio: for every unit of generation work (level 1), the team should reserve between one third and one fifth of that capacity for review and integration (level 3). A human review time ratio per AI task of 1:3 to 1:5 is considered healthy.
Common mistake
Filling the sprint only with level 1 tasks without planning the R&I tasks. The sprint collapses when all the code is "generated" but no one has been able to review it. This pattern produces increments that cannot be considered "done" at the end of the sprint because they have not passed the reinforced DoD, which cancels out the speed benefit that AI provides.
Resources
🏦 AI applied to agile workSkill Arena · Scrum Manager
🏦 Scrum in teams with AISkill Arena · Scrum Manager
📊 Teaching guide: Scrum in teams with AIResources · Scrum Manager
See also
Want to move forward in agility? You can search for course and exam calls or go at your own pace by becoming a member of the Agile Club. This membership includes exclusive resources, e-learning classrooms, and access to Skill Arena: a space to practice and measure your agile skills at your own pace.