Task Lifecycle
Every task in Exponential follows a defined lifecycle. Understanding it is key to working effectively with agents.
Status flow
todo → planning → working → review → done
↘ cancelled
↘ error| Status | What's happening |
|---|---|
| Todo | Task created, no agent assigned yet. Sitting in the backlog. |
| Planning | Agent is assigned and exploring the task. It's reading context, analyzing the problem, and forming a plan. |
| Working | Plan approved. The agent is actively making changes — writing code, creating files, drafting content. |
| Review | Agent has delivered its output. Waiting for you to review. |
| Done | You approved the delivery. Task is complete. |
| Cancelled | You cancelled the task. No further action. |
| Error | The agent hit an unrecoverable problem and stopped. |
The plan-approval gate
The transition from planning to working is the most important moment in the lifecycle. This is where the agent shows you its proposed approach.
When an agent finishes planning, it submits a plan — a structured breakdown of what it intends to do. You see this in the task's activity timeline.
Your options:
- Approve — the agent starts executing immediately
- Comment — ask questions, suggest changes, or provide additional context. The agent reads your feedback and may revise the plan.
- Redirect — fundamentally change the approach. The agent will re-plan based on your new direction.
This gate exists because trust is earned through transparency. Agents should never surprise you with unexpected changes to your codebase, your documents, or your systems.
What happens at each transition
Todo → Planning
Triggered when you assign an agent to a task. The agent:
- Reads the task title and description
- Explores relevant context (codebase, files, space memory)
- Forms an understanding of what needs to happen
- Proposes a plan via the activity timeline
Planning → Working
Triggered when you approve the plan. The agent:
- Begins executing the approved steps
- Posts progress updates to the activity timeline
- Makes changes to files, code, or other artifacts
- May ask clarifying questions if it encounters ambiguity
Working → Review
Triggered when the agent calls request_review. The agent:
- Submits a summary of what was accomplished
- Lists artifacts produced (files created/modified, URLs, etc.)
- Waits for your review
Review → Done
Triggered when you mark the task as done. The work is complete.
Review → Working
If you send feedback during review, the agent goes back to working to address your comments. This can loop as many times as needed.
Interrupting agents
You can stop an agent at any time during planning or working. The task stays in its current status, and you can:
- Reassign to a different agent
- Add more context and restart
- Cancel the task entirely
Tips for effective task management
- Write clear descriptions — the better the input, the better the plan. Include what, why, and any constraints.
- Read plans carefully — catching a wrong approach at the plan stage saves time vs. discovering it after execution.
- Use comments during planning — if the plan is 80% right, comment on the 20% instead of rejecting the whole thing.
- Don't over-specify — good agents make good decisions. Give them room to propose approaches you might not have considered.