Create agent tasks

Agent tasks help you run larger work with a clear goal, selected context, streamed progress, and persistent output history.

Before you begin

• Decide what the task should produce.
• Gather the files or integration records the task needs.
• Choose whether the run may ask clarifying follow-up questions.
• Use a title that helps you find the task later.

Create a task

1. Open Agent Mode.
2. Select New task.
3. Enter a short task title.
4. Write the task prompt.
5. Choose a model such as Fast, Standard, Advanced, Reflective, or Coding.
6. Attach documents, scripts, images, or integration sources if needed.
7. Select topic or persona settings if they are relevant.
8. Run the task.
9. Review streamed progress while the run executes.
10. Open the final output and check it before sharing or exporting it.

Expected result: the run is saved under the task with the prompt, context, status, and output.

Write the task prompt

Use this structure:

Goal: [what the task must produce]
Audience: [who will read or use it]
Sources: [files, ticket, notes, or other context]
Output: [format and sections]
Rules: [constraints, exclusions, or review standards]
Missing information: [ask questions or mark unknowns]

Example:

Goal: Review the attached ISO 27001 access-control evidence and create a gap summary.
Audience: Internal compliance manager.
Output: Markdown table with control, evidence found, gap, risk, recommended action, and owner question.
Rules: Do not claim compliance. Mark missing information as "Not evidenced".
Missing information: Ask clarifying questions if the evidence is not enough to assess a control.

Add files and sources

1. Attach only the documents required for the task.
2. Use Document Manager files for reusable source material.
3. Use integrations only when they are configured and relevant.
4. Confirm that file names are understandable before starting the run.

Tip: For large packs, split the work by document, control, or section so the run can make steady progress.

Use follow-up runs

Use a follow-up run when you want to refine the previous output without rebuilding the whole task.

1. Open the task.
2. Review the latest run output.
3. Add a follow-up instruction.
4. Keep the title unchanged unless the task purpose has changed.
5. Run the follow-up.

Expected result: the task keeps a history of the original run and the refinement runs.

Tips

• Keep names, prompts, and configuration values specific to the task you are performing.
• Check role, subscription, region, and tenant policy when a feature is not visible.

Troubleshooting

Next steps

• Agent Mode overview
• Document Manager
• Build workflows