Create an Agentic AI Agent

Tap into the power of Dialpad’s Agentic AI Chatbot, customer service available 24/7 to not only answer questions, but take action. Agentic chatbots can book appointments, check order status, update records, create tickets, and more, seamlessly connecting to your systems like Bullhorn, Salesforce, ServiceNow, Zendesk, HubSpot, Google Calendar/Workspace, Outlook, AthenaHealth, Calendly, and Zoho CRM.

Let’s dive into the details.

Who can use this

AI Agents are available to Dialpad Support users on an Advanced or Premium plan.

To connect an agentic AI agent to your Contact Center, you must have a Digital Contact Center.

Types of AI Agents

Dialpad offers two types of AI Agents:

• Agentic Agents: Capable of performing tasks, executing workflows, and generating natural answers.

• Generative (Classic) Agents: Designed specifically to generate natural answers from your knowledge base.

This article covers how to create an Agentic AI Agent.

Read this article to learn how to create a Generative (classic) AI Agent.

Create an Agentic AI Agent

To create an Agentic AI Agent, go to your Dialpad Admin Settings.

Note

You must be a Company or Office Admin to create an Agentic AI Agent.

1. Select AI Agents

2. Select Create AI Agent

   

3. Select Agentic

4. Name the AI Agent

   Note

   Your AI Agent’s name is for internal purposes only, customers will not see this name.

5. Enter your Company name

6. Describe your company

   Tip

   The more context you provide, the better.

   Select Improve to use our AI writing assistant tool to help with style and check for errors.

7. Describe the tasks your agent will perform

   

   Note

   This is where you can expand on what you want to agent to do. It can include: how data is displayed, confirmations to ask before proceeding, etc.

   Do not include specific steps for skills. Only include high-level descriptions of the skills you want to agent to perform.

   Select Improve to help craft your prompt.

   Use your discretion when filling this out. The above-listed items are just examples of what you can include in the description for agent behaviour.

8. Define the voice and tone of the agent

   Note

   This is where you can describe the tone it will use, such as a professional tone. You can also define the type of voice that will be used, like active or passive voice.

9. Enter the agent’s display name

   This is the name that will appear to your customers.

10. Describe how the agent will greet your customers.

    Use your discretion when filling out how you want the agent to greet users.

11. Choose the AI Agent’s knowledge base language

12. Select Create

That’s it! Your new Agentic AI Agent is ready for deployment.

Tip

You can update these settings at any time from the Settings section of your AI Agent.

Add knowledge to an Agentic AI Agent

Now that your Agentic AI Agent has been created, it's time to give it knowledge by connecting sources. This could be an internal knowledge base, a public-facing Help Center, or any external website.

Note

Agents have access to the knowledge base throughout all customer interactions and don’t need a specific skill set to use it.

Dialpad’s AI Agents can sync with the following popular content management platforms:

• Zendesk

• Document360

• Box

• Guru

• Google Drive

  Note

  You must select a specific folder, not the entire drive.

  Google Docs, PDF, HTML, TXT, and MD files are supported.

These knowledge connectors sync every 24 hours, ensuring updates to your Help Center articles are automatically reflected in the chatbot’s connected knowledge.

You can also use our custom web crawler to add content from any website.

Not limited to websites, you can also upload helpful documents! Note that documents do not sync with Dialpad, meaning if you update your file, you must remove it and upload the newest version.

To add knowledge to your chatbot, go to your Dialpad Admin Settings.

1. Select AI Agents

2. Select the AI Agent

3. Go to Documents

4. Select Configure sources

   

5. Connect to a content source, or upload documents

   Note

   If you’re connecting to a content management system, you’ll need to provide additional details such as your domain, API key, or locale.

6. Follow the prompts, then select Save

Tip

If you want to upload individual documents, select Add document to add unique files.

*Individual documents cannot be larger than 5 MB.

Agentic Skills

Skills define what an Agentic AI Agent is allowed to do. A Skill grants the agent access to specific actions, for example, adding a caller’s information into your CRM when the caller is a new lead.

These actions are executed through Connectors.

Each API endpoint inside a connector is called a Data Operation. A Data Operation represents a specific “unit of work” the agent can perform, such as Create Contact, Update Lead, or Log Call Note.

Once a connector is published, you can select which Data Operations (actions) and flows you want to make available to your agent by attaching them to a Skill. This allows you to control exactly what the agent can do with that integration.

Skill Prompts and Guardrails

Each Skill includes its own configuration prompts. One of the most important parts of Skill configuration is defining Guardrails.

Guardrails describe the limits of what the Agentic AI Agent can do. They help prevent undesired actions by explicitly defining boundaries—for example:

• What the agent should never do.

• What data the agent should not access or modify.

• When the agent must ask for confirmation before acting.

• When the agent should escalate to a human.

Clear guardrails are essential for safe and reliable agent behavior. While some guardrail enforcement is handled automatically by the platform, specifying them clearly helps ensure the agent behaves as intended.

Create Agentic Skills

To create Agentic Skills for your Agentic Agent, go to your Dialpad Admin Portal.

1. Select AI Agents

2. Navigate to the desired chatbot, then select Manage your agent

3. Select Skills

4. Select Add a skill

   

5. Choose to create a skill from scratch or use one of the pre-defined templates

   

6. If you create a skill from scratch, fill in these fields:

   Skill name	The name of the skill.
   Description	The intended goal of the skill for the agentic agent. This should be kept at a high level of what you want to skill to do.
   Instructions	What you want the agentic agent's skill to accomplish. Be as specific as possible.
   Actions	The action you want to agentic agent to perform. To add an action, type “/” in the field or select the drop-down menu and choose from the list provided.
   Connections	The integration that the Action is linked to.
   Guardrails	Describe what things you don’t want the agentic agent to do within its skill.
   Labels (optional)	Words to describe the skills to enable users to search for a skill performed by the agentic agent.

   

   Note

   If you use a template to create your skill, read through to verify it follows the flow you want. Add guidelines to the highlighted areas to suit your desired process.

7. Once all fields are filled in, select Validate

8. Select Publish

Connectors for AI Agents

Connectors enable the AI Agent to execute the skills you want it to perform. These integrations, available to each AI Agent, enable it to send and retrieve information based on what you want the agent to handle.

To create a Connector, go to your Dialpad Admin Portal.

1. Select AI Agents

2. Select Connectors

3. From the desired integration connector, select Options > Manage Settings

4. If the integration is not enabled, under Enablement, select Enable this feature

   

5. Under Connections, select Add Connection

6. Name the connection and select Connect

7. Select Save changes

To add a Connector to a skill for an AI Agent, go to your Dialpad Admin Portal.

1. Select AI Agents

2. Select the AI Agent you want to add the connector to

3. Select Skills

4. Create a skill to use a connector to a new skill or select Options > Edit to use a connection to an existing skill

   

5. Within the Instructions and/or Guardrails fields, type “/” and choose an action

   

6. In the Actions field on the right-side window, select a connector for the action

   

7. Select Validate, then select Publish

Validating your agent

If optimization opportunities are detected, Dialpad will display suggestions below the Instructions and/or Guardrails sections, suggesting content to add or edit.

If any errors are detected, they are displayed on the right side of the window. You can choose which errors to fix and select Improve.

Note

If you believe the instructions, guardrails, and actions are configured correctly, you can ignore the suggestions and select Validate and Publish > Publish with warnings.

When you have finished adding the suggested content, select Validate and Publish > Validate only. You can continue this process until there are no more content suggestions.

Note

Dialpad may continue to provide content suggestions for instructions and/or guardrails. Use your discretion in how you want your instructions and guardrails worded. You can still publish your agent if there are suggestions provided that you don’t want to include. Be sure to review your content in each section to ensure it’s aligned with your business case before publishing.

Once the content is validated, select Validate and Publish > Publish with warnings.

Test your Agentic AI Agent

Quickly verify your AI Agent’s responses using the Test feature.

To access the Test preview, go to your Dialpad Admin Settings.

1. Select AI Agents

2. Navigate to the desired chatbot

3. Select Test AI Agent

   

4. Enter a question and check your AI Agent’s response

Connect the Agentic AI Agent

Once your Agentic AI Agent has been built and tested, it's time to connect it to your channels so clients can reach your agents when needed.

Connecting your Agentic AI Agent depends on whether you want to use it via a digital or voice channel.

Connect to a digital channel

By default, when creating an agentic AI agent in Dialpad, a new digital channel with the same name as the agentic agent is automatically created and connected to it.

To view the digital channel the agentic agent is connected to, go to your Dialpad Admin Portal.

1. Select AI Agents

2. Select the agentic agent you want to view

3. Select Configure channel

   

4. Under the Info tab on the Digital Channel Settings page, view and/or choose the Contact Center for the agent to be deployed in

   

   Note

   Agent appearance, deployment. general options, and advanced options are also available on the settings page.

To connect an Agentic AI Agent to a different digital channel, go to your Dialpad Admin Settings.

1. Navigate to Channels & Workflows

2. Select Digital Channels

3. Select the digital channel you want to connect the AI Agent to

   Note

   You can also create a new digital channel.

   

4. Under the Info tab, select the Contact Center that you want to connect your AI Agent to

5. Select Save changes

   

Connect to a voice channel

To connect your Agentic AI Agent to a voice channel, go to your Dialpad Admin Settings.

1. Select your desired voice Contact Center, Department, or Main Line

2. Select Business hours and call routing, and navigate to Call Routing

3. Select Edit Call Routing

4. Under Open Hours Routing or Closed Hours Routing, navigate to Routing Options and select No operator

   

5. Under No Operator Options, select To an AI agent

6. In the drop-down menu, select the desired AI Agent

Conversation history

View entire chatbot conversations from the History tab.

Here, you can see how many queries were associated with the conversation, the resolution status of the interaction, the user’s feedback, the number of links that were provided, and the time the conversation occurred.

To view a conversation, go to your Dialpad Admin Settings.

1. Select AI Agents

2. Choose the AI Agent

3. Select History

Tip

Use the search bar to find conversations with specific users, or refine your search using the following filters:

Customer conversations	Filters by customer or test conversations.
All labels	Filters by specific response and document labels.
Unanswered questions	Filters conversations with or without unanswered questions.
Queries	Filters conversations containing no queries or more than 1 query.
Workflow	Filters conversations that did or did not contain a workflow.
Feedback	Filters conversations by with, without, or containing specific feedback.

The date selector in the right-hand corner lets you select specific timeframes.

View conversation details

To view a conversation, select it. On the pop-up, you’ll see conversation details.

• Select Transcript to view the customer’s experience as they spoke with your AI Agent.

• Select Metadata to see the customer’s details, as well as the conversation ID and labels.

Frequently asked questions

How do customers connect to Contact Center agents?

Once the Agentic AI Agent is connected to your digital or voice channel, when a customer says “I want to speak to an agent” they’re automatically escalated into your Contact Center.

What languages does the Agentic AI Agent support?

Agentic AI Agents are available in multiple languages. See the list below.

Can I integrate my Agentic AI Agent into a native mobile app?

Yes. Dialpad’s AI Agent’s native bindings are publicly available for both Android and iOS apps.

If you want to integrate the Agentic AI Agent into your native app, refer to the following public repositories:

• Android GitHub repo

• iOS GitHub repo

Each repository includes:

• A READ ME with setup instructions

• A sample app

• The full framework code

You can use the framework as-is or customize it as a starting point for your own implementation.

My agent monologues on voice, and callers hate it. What’s going wrong?

Long, dense behavior/task descriptions can cause the agent to generate long, essay‑style answers, which are especially painful on voice. For voice, keep behavior prompts short and explicit about step‑by‑step, bite‑sized instructions, and rely on skills and/or the knowledge base for detail instead of writing long explanations in the behavior field.

What should I actually put in the ‘Describe the tasks’ and ‘Describe how this agent should behave’ fields?

Use these fields for high‑level intent and tone, not instructions or guardrails that are already handled at the system level.

• Good content: what types of problems the agent should solve, how formal it should be, whether to split instructions into steps, and when to escalate.

• Bad content: duplicating safety policies, modifying core routing rules, or channel‑specific hacks (“for voice do X, for chat do Y”).

Why is the agent ignoring instructions like ‘Respond in English only’ or ‘Wait 3 seconds before greeting’?

Some instructions contradict how the underlying system works and will be ignored. The model does not have real timing control (“wait 3 seconds”), and core language behavior is usually governed by system‑level prompts and telephony setup, not the per‑agent behavior field. If you see non‑English strings or timing issues, that’s usually a platform bug or configuration problem, not something you should “fix” with agent‑level prompts.

Where do guardrails belong – in my agent prompt, or somewhere else?

Most safety/abuse/self‑harm/malicious‑request guardrails are handled by system prompts and platform‑level logic. Re‑stating them in the agent behavior field tends to confuse the model and bloat the context. Use agent behavior for business‑specific limits (e.g., “Never change a booking within 24 hours of departure; escalate instead”), not generic safety rules.

Why does the bot ‘make something up’ instead of escalating or saying it doesn’t know?

If the agent invents plausible but wrong guidance (“check if your email domain is blacklisted”) instead of escalating. This usually means:

• The fallback policy is too permissive (low confidence threshold).

• There is no clear ‘don’t guess’ rule tied to that topic (e.g., billing, ports).

• There’s no specific skill or knowledge base entry that tells it to escalate.

Treat high‑risk topics as “no‑guess zones”: give them explicit escalation instructions and, where possible, guard them with skills instead of open‑ended answers.