Break work into subtasks and milestones

Big work is easier to watch when it is split. A task can carry child tasks, and any task that has been executed carries a timeline of what actually happened.

The Subtasks tab

Every child of a task is listed on its Subtasks tab, with the count on the tab itself. Each row gives the child’s title, its status, who holds it, and when it is due, and links to the child’s own workspace.

A child is an ordinary task. It has its own key, its own runs, and its own history, and it appears in the task list alongside everything else. The parent link shows up in the child’s Details sidebar as Parent Task.

Where subtasks come from

Children are created by the work, not by a button

There is no add-subtask control in the workspace. Children appear when something creates a task that names a parent, which in practice means one of the three cases below.

• A specialized task reaching its next stage. A content task entering verification spawns the verification child, entering human review spawns the approval, and so on. That is how the pipeline is built.
• A run needing permission. When a run has to be approved before it starts, the approval it raises is created as a child of the task, so the decision is attached to the work rather than floating loose.
• Anything that creates a task with a parent, including work merged from a meeting where one item is plainly part of another.

The milestone timeline

Milestones in the task header opens the execution timeline: what the agent planned, what it did, and what it produced. The count on the button is the number of events recorded.

The page holds four things:

• The plan. The steps the agent laid out, in order, each marked as pending, running, done, or failed, with the tools a step is allowed to use.
• Events. Everything that happened, newest last: plan generated, step started, step completed, step failed, step paused, step approved or rejected, tool calls, model calls, artifacts, and budget notes.
• Artifacts. Files the run produced, downloadable from the sidebar.
• Budget. What the run has spent against its soft cap and its hard cap. Crossing the soft cap is a warning; the hard cap stops the run. See budgets.

A task that has not been executed shows an empty timeline rather than an error. Runs, attempts, and outcomes live on the Execution tab instead; the timeline is the inside of a single agent’s work.

Which one to use

Question	Where to look
Is this work split into pieces, and where is each?	Subtasks tab
What did the agent decide to do, step by step?	Milestones
How many attempts has this taken, and what did they cost?	Execution tab

Next steps

• Work in the Task Workspace: the tabs around these two views.
• Runs and attempts: the execution model behind the timeline.
• Budgets: the caps the timeline reports against.