Every piece of work,
fully accounted for.
A task is the basic unit of measured work. It carries its own cost, context, workflow position, and progress — updated in real time as work happens. No manual status updates, no chasing people for numbers. Every task shows exactly where it stands.
The agent is implementing. Cost and progress update live as it works.
A task isn’t something you write. It’s something your plans produce — already briefed, already tracked.
Someone writes a ticket, tags it, estimates it, grooms it. Status is updated by hand, or not at all. Requirements get pasted into a description and rot. The board is a snapshot from last standup, already wrong.
You think in blocks of work. Spaces generates the tasks from your plans — each one pre-loaded with context, dependencies, and acceptance criteria. Cost, progress, and workflow position keep themselves current as humans and agents work.
Built-in accountability. No status meetings.
Every task updates itself — cost, progress, workflow position, and ownership. You see the full picture without chasing a single update, and it all streams into your dashboards and cost attribution.
Context & attachments
Specs, docs, Figma files, and links live on the task. Agents read this first — you control exactly what they know.
Workflow position
Where the task sits in its pipeline — implement, review, test — with per-step ownership for each human or agent.
Live metrics
LLM cost, iterations, elapsed time, and progress update as work happens. Over budget flips status automatically.
Subtask rollups
Parents aggregate cost and progress from children. The top-level picture stays accurate without drilling down.
Dependencies
What blocks this task and what it unblocks — visible on the card so upstream and downstream stay obvious.
Activity log
Timestamped claims, pushes, cost milestones, status transitions, and comments from humans and agents.
The task is how you brief the agent
Attach specs, design files, API docs, and references directly to the task. When an agent picks it up, it reads everything first. No ambiguity about requirements, no context lost in Slack threads. The task is the single source of truth for what needs to happen and why.
You decide exactly what the agent knows — and it never improvises beyond it.
Hand off between people and agents, mid-task
Every task — and each step in its workflow — can sit with a human, an agent, or hand off between them. Implement with an agent, review with a person, test with an agent: the task records who did what.
The activity log captures both — timestamps for people and agents alike.
Break broad work into trackable subtasks
Agents break broad tasks into focused subtasks automatically. Each subtask tracks its own cost and status. Everything rolls up to the parent — so you see the aggregate without losing the detail. When one subtask stalls, you see it. When costs spike, you see where.
Your whole task list in one view
Filter by status, assignee, plan, workflow step, or cost. See everything across projects and assignees. When a task is over budget or blocked, it’s immediately visible — no digging through boards.
And your agents can work off the same list. Point an agent at it and it works straight through — picking up tasks and running them on its own.