AI Agents (Cognigy) for Process Automation (Task Assist)

AI Agents (Cognigy) for Process Automation (Task Assist) lets agents run predefined, guided business workflows directly inside Copilot for Agents — without switching between systems or screens. Workflows are built and managed in AI Agents (Cognigy) and made available to agents through Copilot.

A workflow can be triggered manually by an agent, recommended by the system based on the conversation, or run automatically through rules configured by an administrator. When a workflow runs, agents are guided through the task step by step — entering information, confirming details, calling connected systems such as CRMs or databases, and completing multi-step processes in one place. When the workflow finishes, it returns a completion status to Copilot indicating whether the process succeeded or failed. Common uses include:

  • Routing tickets to the right team

  • Scheduling appointments

  • Updating customer information

  • Handling escalations

  • Creating and submitting service requests

What Agents Experience

When a workflow runs, the agent sees a series of on-screen forms (cards) that walk them through the task. The workflow can collect information, validate it, take actions in connected systems, and guide the agent from start to finish. Workflows can be:

  • Single-step or multi-step — from a quick action to a full guided process.

  • Multi-form — collecting different information at each stage.

  • Logic-driven — adjusting what happens next based on what's entered.

How a workflow starts

A workflow can be triggered in three ways:

Good to know (current release)

  • AI Agents (Cognigy) is the workflow engine. Workflows are designed and managed in AI Agents (Cognigy), then made available through Copilot.

  • No Virtual Agent Hub required. AI Agents (Cognigy) workflows attach directly to a Copilot Profile.

  • Multi-step, multi-form workflows are fully supported.

How it fits together (Prerequisites)

Setting up AI Agents (Cognigy) for Process Automation (Task Assist) involves two halves that work together:

where

What it does
AI Agents (Cognigy) Where you build the workflow — the forms the agent sees and the logic that runs between them.
Copilot for Agents Profile ( AI Studio) Where you publish and control the workflow — making it available to agents, setting how it's triggered, and managing form data and timeouts.

The order matters, build the workflow in AI Agents (Cognigy) first, then attach and configure it in the Copilot Profile. The profile can only add a workflow that already exists in AI Agents (Cognigy).

Before you begin make sure you have:

  • Access to the AI Agents (Cognigy) platform.

  • Access to NiCE CXone Admin (AI Studio).

  • AI Agents (Cognigy) for Process Automation (Task Assist) enabled for your tenant. If it isn't, contact your NiCE CXone administrator or support to enable it.

How to Setup Workflow in AI Agents (Cognigy)

This section creates the workflow itself. You'll create a project, build the flow, connect it, and test it.

Create a project

  1. Log in to the AI Agents (Cognigy) dashboard.

  2. Under Your Projects, select + to create a new project and enter a Project Name and Project Type (Intent-driven AI or Empty Project).

  3. For channels, select NiCE CXone.

  4. (Optional) Add Skills if your use case needs them. This step is not required.

Build the flow

  1. Go to Build > Flows > New Flow.

  2. Design the workflow for your business process — the forms the agent will see and the steps between them.

Two kinds of forms. A workflow can show forms in two places:

  • In-flow forms — the forms agents fill in during the workflow, built directly in your AI Agents (Cognigy) flow (on the flow's message nodes). These drive the step-by-step experience and can carry information from one step to the next. In-flow forms can also fill in automatically from the conversation and from what is already known about the customer. For details, see Fill In-Flow Forms Automatically.

  • A starter form (optional) — a single form defined in the Copilot Profile that opens when the workflow launches. Use it to capture key details up front, and to pre-fill known information automatically. For details see, Enable AI Agents (Cognigy) for Process Automation (Task Assist).

As you build, keep two requirements in mind:

  • Carry information forward. When a workflow collects information that a later step needs, store it so it's available throughout the workflow. Use the Add to Context node for this.

  • End every path with a status. Every branch of the workflow that returns control to Copilot must finish with a Send Status to Copilot node, set to either Success (the process completed) or Failure (it could not be completed). This is required for the back-and-forth interaction between Copilot and AI Agents (Cognigy); a branch that doesn't end with a status leaves the workflow waiting.

Fill In-Flow Forms Automatically

You can set up a form in your AI Agents (Cognigy) flow so that Copilot for Agents fills in its fields for the agent. Fields fill in automatically as the conversation goes on. The values come from the conversation and from information that is already known about the customer and the interaction.

You turn this on for each form, inside the form in AI Agents (Cognigy). You do not change anything in the Copilot Profile.

The agent stays in control. Copilot for Agents never changes a value the agent entered or changed, and the agent still submits the form. To learn what agents see, see AI Agents (Cognigy) for Process Automation (Task Assist) in Agent Workspace.

To fill in a form automatically:

  1. In your AI Agents (Cognigy) flow, find the form you want to fill in automatically. Forms use the Adaptive Card format.

  2. Make sure each input in the form's body has an id. For example, customerName or incidentDate. You use these IDs in the next steps.

  3. At the top level of the card, next to body and actions, add a metadata object. Inside it, add a fieldsToFill list.

  4. In fieldsToFill, add one entry for each field you want filled in. Set these properties in each entry.

    Property Details
    name Required. The id of the matching input in the card body. It must match exactly. If it does not match, the field is not filled in.

    description

    Required. What the field holds, in plain words. Copilot for Agents uses it to decide which field a customer's answer belongs to.
    dataType Required. The input type of the field: Input.Text, Input.Number, Input.Date, Input.Time, Input.ChoiceSet, Input.MultiChoiceSet, or Input.Toggle.
    customerQuery Required. The type of value to capture: string, number, date, time, array for a field where the agent can select more than one option, or boolean for a field that is on or off.
    questionId Optional. A unique ID for the question. You can leave it out.
    Property Value
    name incidentDate
    description The date the customer's card problem happened.
    dataType Input.Date
    customerQuery date

  5. Leave the rest of the card as it is, and save the card in the flow.

    Use the description to keep sensitive data out of a field. For example, tell Copilot for Agents to capture only the last four digits of a card number, and never a full card number, card security code (CVV), PIN, or password.

As the customer talks, Copilot for Agents fills in empty fields automatically, within a few seconds, when it finds a match. It can use information passed in when the interaction starts, answers from forms completed earlier in the same workflow, and what the customer says. Fields with no value stay empty, and the agent can fill them in. The agent does not need to accept or apply the values.

If a field can get a value from more than one place, Copilot for Agents uses this order, from highest to lowest:

  1. Customer profile data.

  2. Custom JSON data sent from your Studio script.

  3. Values from earlier workflows in the same interaction.

  4. What the customer says in the conversation.

If you do not use a Studio script, for example on Engagement Hub (Multi-ACD/Open), there is no custom JSON data, and Copilot for Agents uses the other three sources in the same order.

A value the agent enters always stays. If the customer corrects something they said earlier, the agent updates the field.

Connect the endpoint

  1. In AI Agents (Cognigy), go to Deploy > Endpoints.

  2. Select New Endpoint and give it a name.

  3. From the Flow dropdown, select the flow you built.

  4. Set the endpoint type to NiCE CXone and select Save.

A single AI Agents (Cognigy) flow can be connected to more than one endpoint if needed.

Test the flow

  1. Open Test Chat in AI Agents (Cognigy).

  2. Select your flow and enter sample inputs.

  3. Confirm the workflow responds as expected.

At this stage the workflow runs independently of NiCE CXone. Once it works here, move on to the Copilot Profile to publish it.