LAB 2: - AI Agents

About this lab

Cisco’s AI-powered Customer Experience (CX) strategy is designed to transform the way businesses engage with customers by making interactions more proactive, more self-service driven, and more empathetic. AI is embedded across our CX portfolio to augment both digital and human interactions, ensuring a seamless, efficient, and personalized experience.

AI Agents serve as the autonomous AI front door to your business, providing instant, accurate, and scalable support for routine inquiries. By streamlining processes and reducing manual effort, AI Agents free up human agents to focus on more complex, high-value interactions.

Lab Objective

This lab introduces you to the following concepts: -

Create an AI Agent in the new AI Agent Studio UI.
Train the Agent on a predefined product spec document as well as the store's Return Policy.
Build Agent actions and supporting workflows in Webex Connect to handle the following:
- Check order delivery status
- Change the delivery date
- Add/Edit a "Safe Area" for delivery
- Cancel an order, with escalation prompts from the AI Agent
Integrate the AI Agent into the final flow from Lab 1 by updating the "Manage My Order" branch.
Use JDS (Journey Data Service) within the Agent action flows to capture context and pass it on to the Agent Desktop.

Background

Lab 2 focuses entirely on setting up the AI Agent within the new AI Agent Studio, combining the ability to answer knowledge-based questions with the capability to perform actions using Connect Flows as fulfillment mechanisms.

We will start by exploring the new AI Agent Studio, which brings together action-based and knowledge-based agents into a single Autonomous Agent. From there, we will configure the fulfillment flows required to replicate the actions covered in Lab 1. Finally, we’ll integrate the AI Agent into the broader flow, demonstrating how it enhances user experience and streamlines workflow automation.

To speed up training, we have provided prebuilt Knowledge Bases and supporting materials, which you can find at https://webexcc-sa.github.io/LAB-2334/kb/

These are the Knowledge Bases contained therein that we’ll use for this lab

Product Information
Order and Returns

Prompt Engineering: A Vast and Evolving Discipline

In this lab, we will cover the basics of prompt engineering, the practice of designing and refining inputs to effectively guide AI models toward generating desired outputs. As it applies to our specific use case, prompt engineering involves structuring prompts strategically to improve accuracy, relevance, and consistency in AI responses. While this will provide a foundation, it’s important to note that prompt engineering is a vast and evolving field with many techniques, strategies, and best practices that extend beyond what we’ll explore here.

Our goal is to introduce core concepts and demonstrate practical applications within the scope of this guide, but this is by no means an exhaustive representation of what’s possible. We encourage you to experiment with different approaches, refine your prompts, and explore additional resources to deepen your understanding. AI models respond dynamically to language, and even small adjustments can yield significantly different results. By continuously iterating and testing, you can uncover new ways to optimize performance and tailor responses to your needs.

Think of this lab as a starting point—there’s a lot more to discover!

Goal 1 – Create an AI Agent in the AI Agent Studio

Open web browser and navigate to https://admin.webex.com and login with the admin credentials assigned to your pod. Under Quick Links Webex AI Agent or under🡪Customer Experience AI Agents.

Note: There will be an error message on the bottom right which can be ignored. It is an error message to let us know that the user was not added to few groups and can be ignored.
The AI Agent dashboard is launched.
Click Create agent button in the top right corner.
Click Start from scratch.
Click Next button.
Click Autonomous and configuration screen will appear.
Fill in the Agent name and Agent’s goal. The System ID is auto generated and the only option under AI engine (right now) is Webex AI Pro 1.0. This is where additional engines can be selected as and when they become available.

Agent name: In the below example, the pod# is 60, so the name is prefixed with Pod60. All the pods AI Agents will be accessible and hence it is strongly recommended to Prefix the AI Agent name with your Pod#.

Feel free to experiment with different wording for the Agent’s goal. You can click on the tooltip link for guidance from the AI product team (this is also where you’ll find the AI Agent product documentation). While we will provide the exact prompts used in our setup, these are not the only valid approaches. There is no single "correct" way to structure a prompt - variations can lead to different outcomes. We encourage you to explore and refine your prompts to see what works best for your use case.

Agent’s goal prompt example:
```
You are a friendly and efficient AI Agent. You are focused on
answering product questions, providing answers to general FAQs and
managing orders for the Cisco Store.
```
We will refine and iterate on the original prompts to demonstrate how adjusting the wording can correct undesired behavior.
Click on Create in the bottom right corner.
The AI Agent profile tab is presented.

Configure AI Agents Instructions

The Agent Goals section defines the overall purpose and role of the AI Agent—what it is designed to achieve and how it should assist users.

The Instructions section provides detailed, step-by-step guidance on how the agent should execute its tasks, including specific actions, response formatting, and behavior rules to ensure accurate and efficient interactions.

You can click Insert Example to view a basic, generic template or refer to the tooltip link for additional guidance. When working with LLM-powered agents, providing clear and actionable instructions is essential to ensure they perform tasks accurately and efficiently.

To achieve this, it’s important to first identify the specific actions the agent needs to perform so they can be explicitly referenced in this section. Based on our goal list, we will define actions for the following capabilities:

Check order delivery status – Action: [fetch_status]
Change the delivery date – Action: [change_date]
Add/Edit a "Safe Area" for delivery – Action: [safe_area]
Cancel an order, with escalation prompts from the AI Agent – Action: [cancel_order]

Our approach will be to provide instructions in a step-by-step format, clearly outlining each action we want the AI Agent to perform. To make these actions explicit, we will enclose them in square brackets, like this: [fetch_status].

Instructions prompt example:

Answer any questions from the user in a friendly and professional
manner, referring to the relevant Knowledge sources as appropriate.
You will also be able to manage any existing orders as follows:  
Collect an Order ID Number ("orderId") to be able to determine what
order is being referenced.

These are your available actions, once the "orderId" is known:

 1. [fetch_status] - use the "orderId" to fetch the status of the
      order.  
 2. [change_date] - change the desired delivery date.  
 3. [safe_area] - add or update the specified safe area for the
     courier to leave the package.  
 4. [cancel_order] - cancels the current order by changing the value
     of "deliveryStatus" to "Cancelled". This action depends on the value
     of "deliveryStatus" not being equal to "Shipped".

 - If an order cannot be canceled because it has already shipped, the
    only way the user can proceed is by contacting a human expert on the
    number \< **WEBEX CONTACT CENTER FLOW ENTRY POINT NUMBER**\>. This is
    facilitated in the \[cancel_order\] action.

 - If the user requests for order status update, provide the order
   status which includes "deliveryStatus" and "deliveryETA". If the
   "deliveryStatus" is not equal to "Shipped", inform the user with the
   response that the order is being processed and they will be informed
   with a delivery date after it is processed. When you respond to this
   message always greet the user in a polite way and include the
   "firstName" in the greeting along with the "orderId".

Note 1: The Webex Contact Center Flow Entry Point Number is the Telephone Number that was assigned to you along with the Pod assignment.

Instructions serve as a guide for the full range of the AI Agent’s capabilities and also provide a space to refine responses retrieved from the Knowledge Base.

Click the Save changes button. This will save the agent and put it into a state that’s ready to be previewed. That completes the initial prompt setup.

A screenshot of a chat AI-generated content may be incorrect.

Goal 2 – Add Knowledge base and training the agent

We will now add knowledge base articles to train the AI Agent.

Click on the knowledge icon in the left side menu, and then click on the “Create Knowledge base” button in the top right of the screen.
Give the Knowledge base a name that starts with your Pod#(number). Ex: - Pod60 Knowledge base and click Create.
You will then see the knowledge base management dashboard where you can add knowledge in one of 2 ways:
File Upload
Documents – This is the method we will use for this lab

Additional means of adding knowledge sources will become available post General Availability, including URLs as sources of information.
Open a new tab in your browser and go to https://webexcc-sa.github.io/LAB-2334/kb/ and copy the Product Information section till you reach Cisco Store. We will copy Cisco store in the next few steps, so keep this browser tab open.
Now tab back to the knowledge base section and click on “Documents”.
Click on “+Create Document” and in the pop for Document name add the following details:
- Document name – Product Information
- Click on “Add to new category” and New Category name – Product.
- Click "Save".
In the Enter Document content section, click on the "Enter document content" to paste the information copied in step #4 and click save on the top right.
Refer to step #4 (open https://webexcc-sa.github.io/LAB-2334/kb ) to copy the rest of the information for Cisco Store: orders and returns.
Now tab back to the knowledge base creation and click on Create Document.
For the document name – Cisco Store Orders and Returns and Add to existing category of Product and click Save.

These files may take some time to save and in the mean time we will continue to next tasks.
Head back to the Dashboard and click on the previously created AI Agent.
Click on the Knowledge tab
Select the Knowledge Base that belongs to your pod from the dropdown list. Note: - Please select your own pod knowledge base since it will display the knowledge base for all Pods.
Click on Save Changes to make the changes “live”.

Note: - You only need to Save your changes to preview them in the Studio. Publishing is only required when you want to push updates into production for use within digital or voice flows.
Go ahead and preview your agent by clicking on Preview.
Type in any question and observe how the AI agent responds. Feel free to reference the product document you uploaded to find some sample questions. Try questions related to both the product details and returns process to verify that the agent can maintain context across both documents.

Goal 3 – Build Agent Actions and Supporting Workflow

Now that we have an AI Agent capable of answering questions from the Knowledge Base, we need to enable it to perform the specified actions.

We’ll show you how easily the Agent can be enhanced using flows in Webex Connect. This approach allows us to achieve the same outcomes as in Lab 1, but with the added efficiency, user-friendliness, and ease of setup that an LLM-powered agent provides.

Let’s begin by navigating back to Webex Connect and opening the service where you created your flows previously.

Step 1: Create the flow in Webex Connect

Login to control hub at https://admin.webex.com and launch Webex connect quick link.
Find the service that belongs to your Pod and click on it to get started.
Go to flows and click on Create Flow.
We have provided a starter flow that gets executed based on the action called within the AI Agent. In this step we will create a flow that is a copy of this starter flow provided. Give the new flow a name prefixed with your Pod#, for ex:- Pod60 All Actions flow. For the method, choose “copy from existing flow”. Click on select flow and select “All_Action_Starter_Flow” and then click “Create”.
In the Configure AI Agent Event window, click save.
Open the “Evaluate” node within the flow to modify.
Modify the text to replace WXCC TELEPHONE NUMBER with the Telephone number that was assigned to the pod. Ex: - if the telephone number assigned was 14082221111, the below text will look like
```
var Outcome = "Order already shipped. The only way the user can
proceed is by contacting an expert on the number +1408221111";
("end")
```
Click save to save the Evaluate node config.
Click Make Live to put the flow in production.
In the next window, click “make Live” again to make the flow live.

Step 2: Create the Action in AI Agent Studio

We will now link the AI Agent instruction of [fetch_status] to the action/task flow that we just built.

Back in the AI Agent menu within the AI Agent studio, click on the Actons tab to open the list of available actions.

By default, there is a pre-created action called Agent Handover. If you do not want the agent to escalate interactions to a human agent, you can toggle this off.

The rules for escalation are entirely determined by your prompting. For example, an instruction like: "When a user asks to speak to an agent, transfer them to a human agent." is enough to trigger the endpoint in the associated Webex Connect or Webex Contact Center flows.

For now, we’ll focus on configuring actions and will also introduce an alternative handover mechanism.
In the Action tab click on New Action to get started.
The first action is called “fetch_status”. Enter the action name exactly as fetch_status, without square brackets. The action name should match the AI Agent instructions provided in Configure AI Agents Instructions.

Enter the following as the Action description:
```
Using the "orderId", fetch the current status of the user's order.
```
Under “Action scope”, select “Slot filling with fulfillment” to indicate that a flow will be used to facilitate the action. This ensures the AI Agent retrieves the necessary information before executing the request.
That will open the “Input entities” list, click on “+New input entity”.

Step 3: Configure the Input Entities

We will now define the entities. Input Entities are the values that the AI Agent will send to the Webex Connect flow, forming the body of the JSON payload you defined in the first node of the flow setup.

For reference, below is the image of the input that we setup in the action flow.

A screenshot of a computer AI-generated content may be incorrect.

These entities represent the variables the agent needs to collect from the user. Therefore, Entity Names are important—they should match the variable names used in your JSON payload to ensure proper data transmission.

Set the Entity Name to “orderId”
For orderId, since we are not setting any constraints, set the Entity Type to String.

Here is the Entity Description we used:
```
The "orderId" is used to fetch the user's order status from the
connected Order Management Platform.
```
Since orderId is needed for retrieving the customer record, it must be marked as a required value.
While Entity Examples are optional, they can be helpful in guiding the AI Agent by providing sample inputs that illustrate what user responses might look like. This can improve the agent’s ability to recognize and extract the correct values from user interactions.
Click Add once you’re done and you’ll see the Entity appear in the list as follows:
We will now add a second entity to pass the action name which will be used in the action flow to branch accordingly. Click on +New input entity to add a new entity.
Set the Entity Name to “actionName” (Note:- pay attention to case sensitiveness).
Set the Entity Type to String.

Set the Entity Description as below:
```
The value is set to "fetch_status" when this action is called.
```
Select “Yes” to mark this entity as required value. Provide entity example as “fetch_status” and click Add.

Step 4: Tie the Action to the Flow

We will now link the “Action” of AI Agent to the flow. Scroll down to the “Webex Connect Flow Builder Fulfillment” section.
- For “Select service” – select your service. The service is named Pod #(your pod number).
- For “Select a flow” – select the flow for Pod# All Actions Flow.
- Click Add at the bottom to complete.
- The action we just added should show up in the list now.
Click Save/Publish
You’ll then see a pop-up where you can assign a Name to the Published Version
Once published, preview the AI Agent to test the action and ensure it is working as expected.
We can now preview and test the AI Agent with actions.
- Try asking for the order status in different ways to see how the AI Agent responds. This will help you assess how well it understands various phrasings and user intents.
- If needed, you can adjust the action descriptions to fine-tune the agent's behavior. Experimenting with different prompts and tweaks will help optimize the agent's performance.
- Since the agent still has access to the Knowledge Base, feel free to mix Q&A-style questions with action-based requests to test the complete user experience.

Congratulations! 🎉

You have successfully created an AI Agent capable of both answering questions from a client data source and executing actions via API calls through Webex Connect.

To build a fully functional agent, you will need to repeat the process for each additional action by following these steps:

Create the Action in AI Agent Studio
Configure the Input Entities
Preview and Test

We won’t be covering the setup of the remaining actions in the same level of detail. Instead, we’ll provide a high-level overview of the completed state for each step of every action.

Goal 4: - Build all the Actions in AI Agent Studio.

We will now build all the actions and link the flow that we built in the earlier section.

Change Delivery Date Action

Go back to AI Agent Studio and this time let’s explore another way of launching the studio. While you are in the WxConnect, go to App tray which is represented as in the tool bar and click AI Agent Studio to launch the studio.
Click on the AI Agent that belongs to your Pod which was created in the section Goal 1 – Create an AI Agent in the AI Agent Studio
In AI Agent configuration, click on “Actions” to configure a new Action.
When you are in AI Agent configuration, click “+ New action”.

Configure the new action with the following details:

Action name – change_date

Action description

    Collect the orderId (if not already known) and ask the user for
    their new preferred delivery date. The delivery date will be used to
    overwrite the "deliveryETA" value in this action. Note that the user
    can change the delivery date, even if the order has already been
    shipped

Action scope – slot filling and fulfillment

Let us now add a new Entity to collect the new_deliveryETA. When the user requests to change the delivery date, the AI Agent asks the user to provide a new delivery date which will be stored in this Entity. Click “+ New input entity” under Slot filling.
Enter the following details for the new input entity
- Entity name: new_deliveryETA
- Entity type: String
- Description:
```
  The new preferred delivery date. This will replace the
 "deliveryETA" value.
```
- Entity examples: you can add some examples if you prefer which allows the AI Agent to train on the format. Ex:- April 4th, April 4th, 2025 etc
- Required: Yes
- Click Add
Let us add the orderId entity in case the order ID needs to be collected. Click “+ New input entity”.
- Entity name: orderId
- Entity type: String
- Description:
```
The orderId of the order to be updated.
```
- Required: Yes
- Click Add
Let us add the actionName entity to set the action name to execute in the flow when this action is invoked. Click “+ New input entity”.
- Entity name: actionName
- Entity type: String
- Description:
```
The value is set to “update_date” when this action is called.
```
- Entity examples: update_date
- Required: Yes
- Click Add
Now we will link the flow to this action under the section Webex Connect Flow Builder Fulfillment.
- Select service: The service that belongs to your Pod. Ex: - if the pod assigned to you is 60, then the service is Pod60.
- Select a flow: Select the Pod# All Actions Flow that was created Step 1: Create the flow in Webex Connect section.
Click Add to add the new action.
We can test this action by clicking “Preview”.
It’s in testing that we may see an interesting response from the Agent:

Sometimes, the agent may demand a date format from us, which is undesirable.

We want to ensure that our AI Agent is flexible in accepting various date formats from users. Since we did not impose a strict format for date collection, this is a great opportunity to fine-tune our prompt for a more user-friendly and desirable outcome.

Go back to the Agent Instructions and add the following to the [change_date] action:
```
*“The user can provide this date in any way, using natural language, and
you will be able to interpret it correctly. For example, if they say
"tomorrow", then you will understand that to mean tomorrow's date. You
will then save this value as mm/dd/yyyy.”*
```
This adjustment will result in a more user-friendly response and improved handling of date inputs.

When testing the PATCH action, be sure to confirm the update by requesting the order status again to verify that the change was successfully applied.

Change Safezone Action

We will now add safe_area action.

In AI Agent configuration, click on “Actions” to configure a new Action.
When you are in AI Agent configuration, click “+ New action”.

Configure the new action with the following details:

Action name – safe_area

Action description

        Collect the orderId (if not already known) and use it to update the
        "safeLocation" value.

        The user can specify or change the location where they'd like the
        courier to leave the package when it's delivered to their home.
        Typical options may include the following:

        1. Front Door  
        2. Back Door  
        3. Garage Door  
        4. Neighbor's House  
        5. Other (please ask to specify)

        Offer the user these options as a numerical list so that they can also
        reply with a number to correlate to their preferred safe location.
        When a number is entered, you'll know this to refer to the desired
        location and you'll update the record appropriately.

Take note of how we’ve structured the prompt to have the AI Agent present options as a numbered list. This approach makes the interaction more user-friendly, especially for SMS-based exchanges, where single-digit responses improve the overall user experience.

Action scope – slot filling and fulfillment

Let us now add a new Entity to collect the new_safeLocation. When the user requests to drop the package at a particular safe area, the AI Agent stores in this Entity. Click “+ New input entity” under Slot filling.

Enter the following details for the new input entity

Entity name: new_safeLocation
Entity type: String

Description:

    The new location that the user would like to have the
    package left when delivered to their home. Please offer the numerical
    list as described in the "action description", along with the option
    to specify any safe location they would like.

Required: Yes
Click Add

Let us add the orderId entity in case the order ID needs to be collected. Click “+ New input entity”.
- Entity name: orderId
- Entity type: String
- Description:
```
 The orderId of the order to be updated.
```
- Required: Yes
- Click Add
Let us add the actionName entity to set the action name to execute in the flow when this action is invoked. Click “+ New input entity”.
- Entity name: actionName
- Entity type: String
- Description:
```
The value is set to “safe_area” when this action is called.
```
- Entity examples: safe_area
- Required: Yes
- Click Add
Now we will link the flow to this action under the section Webex Connect Flow Builder Fulfillment.
- Select service: The service that belongs to your Pod. Ex: - if the pod assigned to you is 60, then the service is Pod60.
- Select a flow: Select the Pod# All Actions Flow that was created Step1: Create the flow in Webex Connect section.
Click Add to add the new action.
We can test this action by clicking “Preview”.

Cancel Order Action

We will now add cancel_order action.

In AI Agent configuration, click on “Actions” to configure a new Action.
When you are in AI Agent configuration, click “+ New action”.

Configure the new action with the following details:

Action name – cancel_order

Action description

Collect the orderId (if not already known) and use it to update the
"deliveryStatus" value. The action will first perform a check to see
what the order status is (if not already known) - if the delivery
status ("deliveryStatus") value is "shipped", then the order can't be
cancelled and the user needs to be put in touch with a human expert on
the number <INSERT YOUR OWN WXCC TELEPHONE NUMBER> THAT WAS ASSIGNED
TO YOU>**. However, for any other status, regardless of what it is,
the order can be cancelled.

Note:- Modify the above text to replace <INSERT YOUR OWN WXCC TELEPHONE NUMBER> with the telephone number assigned to your pod.

Action scope – slot filling and fulfillment

Let us now add a new Entity to collect the orderId. When the user requests to drop the package at a particular safe area, the AI Agent stores in this Entity. Click “+ New input entity” under Slot filling.
Let us add the orderId entity in case the order ID needs to be collected. Click “+ New input entity”.
- Entity name: orderId
- Entity type: String
- Description:
```
The orderId of the order to be cancelled.
```
- Required: Yes
- Click Add
Let us add the actionName entity to set the action name to execute in the flow when this action is invoked. Click “+ New input entity”.
- Entity name: actionName
- Entity type: String
- Description:
```
The value is set to “cancel_order” when this action is called.
```
- Entity examples: cancel_order
- Required: Yes
- Click Add
Now we will link the flow to this action under the section Webex Connect Flow Builder Fulfillment.
- Select service: The service that belongs to your Pod. Ex: - if the pod assigned to you is 60, then the service is Pod60.
- Select a flow: Select the Pod# All Actions Flow that was created Step1: Create the flow in Webex Connect section.
Click Add to add the new action.
We can test this action by clicking “Preview”.

With the default value set to “shipped”, the AI Agent will always inform the user that the order cannot be canceled and will provide the appropriate contact number for further assistance.

This mechanism also serves as a built-in escalation path to a human agent, entirely handled through prompting, without requiring additional flow logic.

Verifying the Action Flow:

If you want to test the cancellation process, use Update CRM (The process can be found in Lab 1 guide) to manually set the deliveryStatus variable to any value other than “shipped”.

Once updated, the AI Agent will recognize the change and proceed to cancel the order, updating the status to CANCELLED accordingly.

Please click “Publish” to publish the AI Agent and put it in production.

That concludes Agent Actions and the AI Agent Studio portion of this lab. 🚀

Next, we will focus on simplifying the final flow from Lab 1 to:

Integrate the AI Agent to streamline the building process

Now, let’s head back over to Webex Connect to continue!

Goal 5: - Integrate the AI Agent into the flow.

At the end of this goal, you will release the importance of AI Agent in simplifying the workflows. It allows the brands to move from a structured workflow which requires you to capture and program all paths of interaction to meet the required customer experience to have an unstructured flow with AI Agent. It is also important to understand that AI Agent is not to be deployed once and forgotten. It is important to review the sessions to understand the conversation failures and enhance the instructions to add the required guardrails or modify the language to meet the needs.

We are now going to modify the flow we built in Lab 1 and incorporate Ai Agent into this flow. In order to use the time efficiently, there is a Lab 2 starter flow provided in your Pod service, and we will use this flow to copy a new flow and change few settings before putting into production.

Login to control hub at https://admin.webex.com and launch Webex connect quick link.
Find the service that belongs to your Pod and click on it to get started.
Go to flows and click on Create Flow.
In create flow window enter the following details:
- Flow Name: Pod# Lab2 AI Agent (substitute # with your Pod number. Ex: - in the image below, the Pod# is 60 and hence the flow is named Pod60 Lab2 AI Agent).
- Method: Copy from existing flow.
- Select Flow: Lab 2 starter flow.
- Click “Create”.
Configure webhook window appears. We are going to create a new webhook for Lab2.
- Under Configure webhook settings to trigger this flow
- Create new event.
- Name – Pod# Lab2 AI Agent Trigger Ex: - For Pod 60, name is Pod60 AI Agent.
- Example Data: Use the below JSON payload.
```
{

"orderId": ""

}
```
- Click Parse to parse orderId as a variable.
- Click Save.
We now need to modify the SMS nodes to add the sms number. In the below flow, there are three SMS nodes identified. Open the SMS node marked as 1 by clicking on it.
The SMS configuration window opens and check the “From Number”, it is blank, and we need to fill this with the SMS number assigned to your pod.

Click on the “From Number” and select “yourAssignedSMSNumber” variable from the drop down under Custom Variables and click “Save”.
Repeat the steps to complete the configuration for SMS nodes #2 and #3.
Next, we will verify the Receive node is configured. There is only one Receive node in the flow that is highlighted. Open the receive node and click save even if the node is configured correctly.

If the Receive node Number is blank, click on the Number field and select “yourAssignedSMSNumber” variable from the Custom Variables drop down. Select “*” for the keyword and click Save.
Next, we are going to modify the AI Agent nodes in the flow. There are two AI Agent nodes in the below flow that needs to be modified.

Open the AI Agent node #1 to modify. Click the AGENT field and in the drop down pick the AI Agent name that we configured in “ Goal 1 – Create an AI Agent in the AI Agent Studio”. Ex: - The pod assigned below is 60, so the AI Agent picked is Pod60 AI Agent.

Click Save to complete.
Repeat the steps for AI Agent node #2, the completed node configuration of AI Agent node #2 is shown below. Click Save.
All the configuration changes for this flow are complete. Click “Make Live” to put the flow into production. The Make Live configuration window appears as shown below.
Click Make live again to put the flow in Production.

Goal 6 – Testing the AI Agent

It is now time to test the flow that we put in production. The flow that we built is triggered by certain notifications from the CRM or system of record updates as a proactive notifications. We will simulate this by creating this notification and keep your mobile phones ready, let us begin testing.

Note: - If the delivery status is set to “Not Shipped”, use the “Update CRM” to set the delivery status to “Shipped” for this part of the testing and do the order cancellation as the last step, as this will be the lead into lab3.

Open the flow that we put into production and double click on the first node “Trigger AI Agent Flow”.

In the configuration window, locate the field named “Webhook URL” and the url is of the format https://hooks.us.webexconnect.io/events/XXXXXXXXX. Copy the last part of the URL “XXXXXXXXXX” to test.
Open a web browser and login to http://crm.cxocoe.us with the credentials provided.
Click on “Kick off Proactive Flow”.
- Phone Number: Enter your mobile number with country code. Ex: - 14081111111
- Webhook Id: Enter the copied text from step #1
Then click on “Start Flow” to trigger the proactive journey.
The flows provide debug capability to identify any issues with the configuration. The debug can be accessed by clicking on bug icon that is located on the right of the flow.

The debug window contains list of transaction id’s that the flow has executed. Click on the appropriate transaction id to debug.

The debug output is encrypted by default. Click “Decrypt Logs” to view the logs in plain text.
You should have received a proactive notification on your mobile phone notifying you the purchase of a Core Trio Qi charger!
The above test is with delivery status set to “shipped”. Optionally, the test can be repeated for delivery status set to “Not Shipped”.

Open web browser and login to http://crm.cxocoe.us with the provided credentials.

Note: In case you do the testing with setting the delivery Status to “Not Shipped”, please repeat the test again with delivery status set to “shipped” and going through the order cancellation as the last step because this will be the lead into the Lab 3 testing.
Click on “Update CRM Data”

The field to update is Delivery Status and the value is “Not shipped”. Then click “Update Record”.
Now let us test again by repeating Step #3
Verify the messages that you received on your mobile phone

Comparison of Structured flow vs AI Agent

Let us know compare the flow that was created for Lab 1 vs Lab 2 to understand the advantages of AI Agent to handle an unstructured conversation and its impact on the brand.

Here is the flow we built for Lab 1

A diagram of a network AI-generated content may be incorrect.

And now below is the flow we built for lab 2 with AI Agent.

A diagram of a computer AI-generated content may be incorrect.

Even though we are able to provide similar brand experience for our customers, AI Agent with its ability to handle unstructured flow can prove beneficial for the brands to service the customers more efficiently and in a timely manner without having to program all the communication paths like in structured conversation.

This is an example of Cisco Autonomous AI Agent which is based on “Generative AI” technology, but if brands prefer to leverage a more structured SelfService AI Agent, Cisco is also positioned to provide this solution using “Scripted AI Agent”.

👏👏Congratulations!! This concludes Lab 2.👏👏