Agent tasks

Agent tasks are how you give Zahen a goal that requires more than a single answer. The agent plans a sequence of steps, executes them one at a time, saves its progress at each checkpoint, and pauses for a human decision whenever a step carries enough risk to warrant one.

Tip

Unlike a quick question — which is read-only — an agent task can take real external actions such as drafting an email or looking up a live record. That power is why every action type carries an explicit risk level and why high-risk steps always pause for approval before they happen.

In this section

• Tasks vs. quick questions — when to start a task instead of asking a question.
• Starting a task — the new-task form, what a good goal looks like, and the intake fields for each assistant.
• Following a task — the execution timeline, live status, citations, and tool calls.
• Risk levels & approvals — low, medium, and high risk; why a high-risk step pauses before it acts.
• Task history — finding and reopening past tasks in both the console and Employee mode.