You can put an AI agent to work in two ways: automatically, through a trigger, or by hand, assigning it as the analyst of a case or an onboarding.

Both paths coexist. The usual setup is an automatic trigger for the day-to-day flow, plus manual assignment for the one-off cases someone decides to hand over.

A trigger tells the agent "when this happens, on this case type, start working". Set them up in Settings > AI & Agents > Agents configuration, inside the agent's "Triggers" section. Every agent needs at least one: with no triggers it would never run, so the platform does not let you save it.

To add one, click "Add trigger" and pick the event and the associated template. The same agent can have several.

The agent starts as soon as the case or onboarding opens, without waiting for anyone to take it.

This is the option for volume work: monitoring alerts coming in all day, screening hits generated automatically, onboardings arriving from the public form. The agent makes the first pass and your team reaches the case with the evidence already gathered.

It fits when the initial analysis is always the same and does not depend on prior human judgment.

The agent starts when someone assigns it as the analyst of the case or onboarding.

This is the most flexible trigger, because it turns the assignment into the start signal. An analyst reviews the case, decides it needs a deeper look, and hands it to the agent. The trigger fires only when the analyst decides so.

The analyst on this trigger is always the agent itself, so that field is fixed when you configure it.

It fits when the agent does expensive or deep work that is not worth running on every case.

The agent starts when the case or onboarding enters a given stage. You also pick the stage when configuring it, and only the template's active stages are offered.

This is the option for slotting the agent into an exact point of your process. For example, having it review the file only once it moves to "Risk analysis", after someone validated the basic documentation.

It fits when the agent needs the case to already hold certain information before it can work.

The agent starts once the customer finished sending all the requested documentation. This event exists only for Kayce.

It is the natural close of the onboarding loop: you request documentation, the customer responds, and the agent analyzes the complete file instead of working with partial data.

To put an agent to work on a specific case, open the case, change the analyst to the agent from the "Analyst" dropdown and save.

This requires the agent to have the "When an analyst is assigned" trigger configured. Without that trigger, assigning it as the analyst does not put it to work.

You can also do it on several cases at once: select the cases in the list with the checkboxes on the left and use bulk edit to change the analyst on all of them.

The agent shows up as the case analyst and works with the permissions you gave it. It leaves a record of what it does: it publishes its report if the "Write report" tool is on, and moves the case status if you enabled those statuses.

Everything keeps the same traceability as any team member's work, so you can audit it step by step.

If you expected the agent to start and it did not, check these points in order:

If you deactivate a stage that has a trigger pointing at it, the platform warns you first and deletes that trigger. A deactivated stage never changes, so the trigger would have no effect.

This is worth checking when you redesign a flow: reordering stages can leave an agent that was working without its trigger.