Run three agent tasks without losing the next owner
Divide the work by independent outputs, not by the number of agents available.
Workflow · Updated 2026-10-20 · 3 min read
Divide the work by independent outputs, not by the number of agents available. Give each output one owner and record the handoff between tasks. Use Tinovio to keep those assignments and results visible across sessions.
This proposed example uses research, implementation preparation, and test planning for a checkout change. The work can begin in parallel only where the inputs are independent. Implementation must still wait for decisions it depends on.
Describe the shared outcome
The checkout must allow purchases without registration and show payment errors clearly. Save that requirement and its reason as shared context. Every task should refer to the same decision.
Name the repository version the agents inspect. If one participant uses a different version, its findings may not apply to the builder's code. Record that difference instead of silently combining results.
Assign three distinct outputs
Create these tasks in the project Space:
- Research: identify the current payment-error path and unresolved questions.
- Implementation preparation: identify the components that the change should preserve.
- Test planning: define observable checks for guest checkout and declined payments.
Each task gets its own agent session and completion condition. Keep the preparation tasks read-only if their purpose is investigation. Three simultaneous editors are not necessary for three useful outputs.
Make dependencies explicit
The test planner can define expected behavior before the code changes. The investigator can inspect the existing response handler independently. The builder should wait when either task reveals an unresolved product decision.
Record dependencies where the task model supports them, and explain the required input in the task detail. Do not call work independent merely because the agents run in separate windows.
Keep findings in the relevant threads
Each owner answers questions in its task thread and places longer material in a shared file. The handoff should identify the artifact, inspected version, and consequence for the next task.
Ask the owners to address the builder when their inputs are ready. The builder then reads the research, component boundary, and test plan together. Shared context provides the common requirement behind all three outputs.
Create the implementation from confirmed inputs
The builder should distinguish confirmed findings from hypotheses in the briefs. It records any new decision before making the change. Then it implements under one build task with a clear code boundary.
If several agents must edit, use separate branches or worktrees and agree how their changes combine. Tinovio's coordination record does not lock the repository or merge changes automatically.
Verify the combined result
Assign a separate review of the fixed implementation version. Give the reviewer the original requirement and the actual test results. A completed preparation task is not proof that the final checkout behavior works.
Check that every completed task has a useful artifact and every remaining task has a next owner. Read the task-board guide when those records are incomplete.