Three-Level Estimation
Unlike traditional estimation in story points or ideal time, three-level estimation does not seek to quantify effort but to classify the type of intelligence each task requires. This classification determines who does what in the sprint and how much review capacity must be reserved.
The three levels
Level 1 — Automated task
A task that an AI agent can execute autonomously with minimal supervision, provided that the spec is complete and the appropriate boundaries have been defined.
Characteristics: clear and formalizable requirements, expected deterministic behavior, no dependencies on organizational context that cannot be documented, and tests that can verify the result.
Examples: code generation from a Given/When/Then spec, scaffolding of components with established patterns, refactoring following defined rules, and generation of unit tests for a documented function.
Level 2 — Human task
A task that requires human judgement, organizational context, creativity, strategic decision-making, or an understanding of the real problem that AI cannot provide.
Characteristics: involves ambiguity that requires human negotiation, affects high-impact architectural decisions, requires an understanding of the real user and their needs, or entails direct legal or ethical responsibility.
Examples: defining the Sprint Goal, prioritizing the backlog, designing the system architecture, interviewing users, making product decisions under uncertainty.
Level 3 — R&I Task (Review and Integration)
Human work on the agent's output: verifying that what has been generated is correct, integrating it with the rest of the system, and making sure it meets the reinforced DoD. This is the R&I task.
Characteristics: always accompanies level 1 tasks, consumes real sprint capacity that must be explicitly planned, and cannot be fully delegated to another agent.
Why the three levels matter in Sprint Planning
Without this classification, teams tend to plan the sprint as if all tasks were level 1 (which the AI generates) and forget level 3 (which humans review). The result: at the end of the sprint there is a lot of generated code but little code that has been reviewed, integrated, and approved. The increment cannot be declared "done".
The rule of thumb: for every hour of level 1 work, reserve between 12 and 20 minutes of level 3. Healthy ratio: 1:3 to 1:5.
Common mistake
Classifying as level 1 tasks that are actually level 2. A task seems automatable until the agent starts generating it and the hidden complexity surfaces: undocumented dependencies, implicit business rules, expected behaviors that nobody wrote into the spec. The team stops the agent, documents what is missing, refines the spec, and runs it again. If this cycle happens frequently, it is a sign that the refinement process prior to the sprint is not properly identifying which tasks are truly ready for level 1.
Resources
🏦 Scrum en equipos con IA [Scrum in Teams with AI, in Spanish]Skill Arena · Scrum Manager
📊 Guía didáctica Scrum en equipos con IA [Teaching Guide: Scrum in Teams with AI, in Spanish]Resources · Scrum Manager
See also
Want to advance 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.