Bringing Scrum to Development with AI

Bringing Scrum to Development with AI

How people and AI work together Putting Scrum into practice through its roles and events Human Clarify the work together Sets direction and priorities. Decides what to adopt. Start when ready Planning Developers → Coordinator Human approves Developers AI Two developers, two separate solutions Product Owner AI Turns decisions into issues Create / update GitHub Issue Development Build and test Review each other's work Review AI PO checks acceptance ↓ Human reviews & decides Retrospective Proposes improvements Human chooses changes AI PO uses findings to update specifications and issues Coordinator AI Manages the entire cycle Hand off findings 2

Who does what Human Stakeholder / PO Sets direction and priorities with the AI PO Makes decisions on plans, results and improvements Coordinator AI Manages each issue and compares the two solutions Merges the chosen work, cleans up and runs the retrospective ⇄ Discuss and clarify PO AI Works through priorities and writes specifications and issues Checks the result against acceptance criteria Developers AI Plan, build and test separate solutions to the same issue Review each other's pull requests (PRs) PO = Product Owner. The coordinator is the template’s “development team” role. There is no separate Scrum Master. 3

Events & activities Refinement Planning Development Review Refinement: defining what to build Human Discusses the goal with the AI PO. Sets direction and priorities. Discuss & clarify ⇄ PO AI Records decisions in the specification (spec.md) and scopes the next issue. Retrospective GitHub Issue Write → Serves as a Product Backlog Item (PBI). States the purpose, scope and acceptance criteria. The PO orders the work and marks an issue ready. This starts the coordinator and planning. Here, the human and AI PO lead refinement. Developers join at planning. In Scrum, developers also take part in refinement. 4

Events & activities Refinement Planning Development Review Planning: agreeing on what and how Retrospective Each issue follows the flow of a Sprint, without a fixed Sprint timebox. Developer AI Claude Code Reads the issue and drafts a separate plan. Developer AI Codex Reads the issue and drafts a separate plan. Two plans → Coordinator AI Presents both plans and open questions to the human. Plans → Human Checks the plans against the goal. Refines and approves. Gaps in the specification? Resolve them with the PO. ↩ Once the human approves, both developers build and test their own planned solution. This process does not include a Daily Scrum. 5

Events & activities Refinement Planning Development Review Development: two solutions, mutual review Retrospective Two developers work independently on the same issue, using the plans the human approved. Developer AI Claude Code Builds and tests the approved plan. Opens PR 1. Review each other's PR ⇄ Developer AI Codex Builds and tests the approved plan. Opens PR 2. Two PRs Coordinator AI and review findings Compares both → solutions and the review findings. The coordinator recommends a PR for the AI PO and the human to review. Within the approved plans, AI handles implementation, testing and mutual review. No human work is required during this stage. 6

Events & activities Refinement Planning Development Review: inspecting the result and deciding Review Retrospective After developer reviews, the AI PO checks acceptance. Then the human reviews the result. Coordinator AI Compares the two solutions and reviews. Recommends a PR. PO AI Proposed PR → Checks the result against the specification and acceptance criteria. No code review here. If changes are needed, the coordinator arranges fixes and another check. ↩ Human Acceptance Inspects the actual result. findings Decides whether to → adopt, request fixes, or revise the criteria. Revising criteria? Update the specification and issue with the PO. ↩ After the human chooses the result, the coordinator merges it, cleans up and starts the retrospective. 7

Events & activities Refinement Planning Development Review Retrospective: improving how we work Retrospective Once the chosen result is merged, look back at how the issue was handled. Coordinator AI Uses test and review records to identify improvements backed by evidence. Proposed changes → Human Considers each proposal. Chooses which changes to adopt. Approved changes → Coordinator AI Updates procedures and documentation. Passes specification findings to the AI PO. The AI PO and the human use the findings to refine future specifications and issues. Check whether the changes help on the next issue. Approving a change, applying it and checking its effect are separate steps. 8