Create an Agent Co-worker

Who this is for. Super Admins creating an Agent Co-worker. See the Roles and Access.

An Agent Co-worker is an instance of an agent, configured through a wizard opened from the Agent Co-workers page. The same wizard edits an existing instance. Only the heading changes: New AI agent co-worker when you are creating one, Edit AI agent co-worker when you are changing one.

Before you begin

• A system of record connection and a connected mailbox must already exist. Neither can be created from inside the wizard.

• Creating an Agent Co-worker requires the Super Admin role. No other role can view or edit the configuration.

• Settle the planning decisions first. For more information, see Before You Start.

The system of record is fixed when the Agent Co-worker is created. It cannot be changed afterwards, so confirm it before you activate.

Open the wizard

1. Open Agent Co-workers from the left rail. Agent Co-workers are grouped by autonomous process, each group showing how many agents it holds.

2. Select the agent the instance belongs to, for example SmartVendor for AP work. The Running Instances popup opens.

3. Select Create Agent.

Figure 1. The Running Instances popup, where Create Agent opens the wizard.

The popup lists the instances that already exist for the agent, with their agent name, ERP, mailbox, status and reply state. Each row carries three actions: Pause, Delete and Edit. Edit opens this same wizard on that instance. Show Description expands the agent description above the table. For more information, see Maintain Agent Co-workers.

If Create Agent is not there, your role does not permit creating instances.

Steps in the wizard

The progress bar shows six steps, but two of them are placeholders in the current release. They stay in the bar, grayed out, and the wizard steps over them.

Table 1. Steps in the progress bar

Selecting Next on Define capabilities goes straight to Activate. Nothing in the wizard exercises the configuration before it goes live.

Step 1: Define the working scope

Step 1 has four fields. It names the agent, and it binds the agent to a system of record and a mailbox. Those two bindings decide what the agent can read and which mail it works. Agent Name is required, takes 3 to 75 characters, and must be unique in the tenant. The wizard checks the name against the agents that already exist and against the names reserved for other system of record and mailbox pairings, and marks the field when the name is taken. Agent Description is optional and takes up to 300 characters. Editing an instance that has no name yet opens both fields filled in from the system of record and the mailbox, and you can change either one.

Table 2. Connection fields on Define working scope

Figure 2. Step 1, Define working scope.

Next stays unavailable until the agent name, the system of record and the mailbox are all set.

The mailbox list offers only mailboxes that are free. A mailbox already bound to an active instance is hidden, so a mailbox you expected but cannot find is already in use. Check the agent’s instance list rather than the mailbox connection. The system of record list is not filtered this way, so several instances can work the same system of record.

Step 2: Choose functions

Step 2 decides which AP functions this co-worker handles. Each function you select gets its own tab on the next step.

Figure 3. Step 2, with both available functions selected.

Select at least one function. The line under the functions counts what you have chosen and tells you when nothing is selected yet. Selecting a function again clears it. Both AP functions are offered in the wizard:

Table 3. Function availability

Anything you leave out falls back to platform defaults.

Step 3: Define capabilities

Step 3 carries one tab per function you selected, and each tab has its own numbered rail of sections down the left. Work through the rail in order the first time. The rail scrolls, so the last sections sit below the fold.

Figure 4. Step 3, with the AP Invoices tab open on Autonomy level.

The AP Invoices tab lists these sections in rail order: Autonomy level, Document intake, Field extraction, Business rules, and Write to SOR. For more information, see the AP Invoices guide.

The AP Helpdesk tab lists these sections in rail order: Autonomy Level, Contact Management, Email Categories, Vendor Categories, Processing & Automation, Processing Rules, Assignee Hold, Email Archiving, and Task SLA. A Mailbox analysis panel sits above them. For more information, see AP Helpdesk capabilities.

Set the autonomy level first on either tab. It governs which of the sections below it apply.

Step 6: Activate

The final step holds a single instruction and the Run button. It shows no summary of what you configured, so any review has to happen before you reach it. For more information, see Review and Activate.

Figure 5. Step 6, Activate. Steps 4 and 5 stay grayed out in the bar, and Run sits at the foot of the page. The empty area between the two is omitted.

Select Run to save the configuration and activate the agent. The button is labeled Run, not Activate.

When it succeeds the browser opens the new instance dashboard, and the instance appears in the agent’s Running Instances list with the status Running.

Activation can fail on the first attempt. Selecting Run a second time, with no change to the configuration, usually succeeds. For the message it returns and what to check if the second attempt also fails, see Review and Activate.

After activation

The Agent Co-worker begins processing inbound email in the connected mailbox. The first days of live traffic are where configuration problems surface, so check which rules are applying, read a sample of agent-drafted replies before they send, and confirm tasks are routing to the people you expected.

Related information

• Before You Start

• Define Capabilities - AP Helpdesk

• Review and Activate

• Maintain Agent Co-workers