Agent Tasks
Queue, prioritize, and track tasks executed by AI agents
Autopilot: 7 pages
Agent Tasks are the unit of work for your AI agents. Each task represents a specific goal for an agent to accomplish — from researching a competitor to drafting a follow-up document.
How Tasks Are Created
Tasks enter the queue from multiple sources:
| Source | Description |
|---|---|
| Playbooks | When a playbook wakes and its evidence rules pass, it creates exactly one task for its worker, carrying the evidence that justified the run |
| Conversational | Chat with Autopilot in its DM channel: "Go research competitor X" |
| Multi-agent delegation | One agent assigns a task to another via assignAgentTask |
| Manual | Create tasks directly from the Task Queue on the Agents page |
Task Properties
| Property | Description |
|---|---|
| Title | Short description of what needs to be done |
| Instructions | Detailed goal text the agent will execute |
| Priority | Low, Medium, High, or Urgent |
| Status | Pending, Queued, Running, Waiting (paused for your input), Completed, Failed, or Cancelled |
| Agent | Which AI agent will execute the task |
Task Execution
Tasks are processed by the Agent Task Executor Job, which runs every 15 seconds:
- Claims up to 3 pending tasks atomically (using
FOR UPDATE SKIP LOCKEDto prevent duplicates) - Executes each task through the Agent Runner, recording its reasoning and tool calls in the task's Activity timeline
- Updates task status and persists results
- For tasks created by a person, posts a "Task completed" summary to that person's DM with the agent in Team Chat
- Memories come from the agent itself: during the run it can call
saveMemoryto keep findings for future context (task output isn't mined for memories automatically)
Avoiding Busywork
Recurring work doesn't come from an agent deciding on its own what to do next. A playbook acts only when its evidence rules pass against your real data — otherwise it records a quiet run and creates no task. See Playbooks.
Conversational Task Management
Chat with Autopilot in its DM channel to manage the task queue:
- "What are you working on?" — See the current task queue
- "Add a task to research competitor X" — Queue a new task
- "Cancel the email outreach task" — Cancel a pending task
- "Focus more on outbound leads" — Save as guidance for future work
To start a playbook's work immediately, open it on the Playbooks page and click Run now. It goes through the same evidence check as a scheduled run.
Viewing Tasks
- Task Queue on the Agents page — every agent task with its status; open one to see its Activity timeline and results
- Agent detail page — Tasks assigned to a specific agent
- Playbooks → Work — every playbook run, including how many tasks it created
Cost Controls
Task execution respects plan-based limits:
- Each run is checked against the agent's token budget and your plan's AI credits before it starts, and its usage is recorded afterwards
- Free and Starter plans are capped at the standard model tier; Pro and higher plans can use the complex tier
- A generous per-run step limit stops runaway loops; it isn't meant to cut normal work short