Skip to main content
An agent is a named AI teammate with a role, context, model settings, and tools available in your workspace. Its role helps you decide what to ask; its capabilities still depend on workspace permissions and connections.

​
Choose an agent

Use the Agents area in Agent Chat, search by name, and select the agent you want to talk with. Check the active agent before sending. Agents can have multiple sessions; choosing an agent does not combine their conversation histories. Use Create agent when you need a reusable teammate with a different role. Use a new session when you need a separate conversation with the same agent.

​
Give enough context

Include the outcome, audience, constraints, source material, and how you will review the result. A useful request is:
For existing tracked work, reference the actual task ID and project. An attached task or project reference helps the agent locate context; it does not move the conversation into another project or approve execution.

​
Owner, assignee, and active executor

The Owner represents task responsibility. The Assignee identifies who is assigned to the task. An active execution records the agent run currently doing the work. These fields have different purposes. The current Task Detail supports one assignee, not a multi-assignee list. Changing task responsibility alone does not transfer an active run. Use the supported lifecycle or handoff process if another agent must continue running work.

​
Follow progress

Use chat for discussion and Task Detail for lifecycle, next action, attachments, and execution history. If work is Waiting or Blocked, provide the missing decision or resolve the recorded obstacle. If you need another agent’s review, request it explicitly and identify the exact output. See Agent management for profile, context, and availability controls.