Skip to main content

Agent Creation

Learn how to create, configure, and share your own AI Agents in watsonx Workshop.

The AI Agent Creation Guide walks you through the end-to-end process of building AI Agents using the Agent Creator experience in watsonx Workshop. Whether you're creating a lightweight assistant or a multi-skill AI Agent with advanced workflows and integrations, this guide provides step-by-step instructions to help you get started.

Learn how to:

  • Create and configure your AI Agent’s core details and overall experience
  • Add Static and Dynamic Knowledge sources to ground your AI Agent
  • Build Basic Agents or Skills Agents for different use cases
  • Configure external APIs, integrations, and agent workflows
  • Enable Studio tools such as Podcasts, Client Role Play, Presentations, and Practice Gym
  • Test, refine, and share your AI Agent across IBM

What You’ll Explore​

Skills Agents & Integrations​

Learn how to build and configure AI Agents tailored to different business and learning use cases. This includes defining agent behavior, grounding agents with the right knowledge sources, configuring prompts and workflows, and designing experiences that align to your intended audience and outcomes.

Dynamic Knowledge & Context​

Ground your AI Agent with real-time and contextual information through:

  • IBM Product Context
  • Industry & Sector grounding
  • Adaptive Context
  • Web Search

Studio Experiences​

Enable interactive and AI-powered learning experiences including:

  • AI-generated Podcasts
  • Client Role Play simulations
  • Presentation creation
  • Practice Gym exercises

Step-By-Step Guide​

Begin the step-by-step walkthrough for building AI Agents in watsonx Workshop. This guide covers essential configuration details, recommended best practices, and practical setup examples to help you design, deploy, and refine your agents effectively. Dive in to accelerate your development process and ensure a smooth, scalable implementation experience.

Step 1: Access the AI Creator Space and Provide Basic Agent Details

Go to your personal Agent Creator space and click Create Agent to get started.

The layout has been designed to be easy for you to create and test your agents as you build. You will configure your AI on the left of the screen, within each of the 4 sections and at any time, you can test your agent by sending messages and seeing the responses on the right of the screen. Testing as you go increases the quality of the agent and increases the value your agent brings to your target users.

Expand the 1. Agent Details section in the top left of the screen. You should now see the screen below.

Provide the basic details of your AI Agent, including:

  • Agent Name: The name that your users will see.
  • Agent Objective: A short description explaining what your AI Agent does for users.
  • Agent Description: A more detailed description of your agent, how it works and what your users should expect.
  • Tags: Short keywords relevant to your AI Agent that help users discover it through search. Separate each tag with a comma. For example, the Build Pipeline with Products Agent would have the following tags: Pipeline, Demand Generation, Create Pipeline, Create Deals, Lead Generation.

It’s important to provide these details clearly and concisely, as they help IBMers discover your AI Agent when starting a new session.

Step 2: Define the Agent Data Sources

Ensuring your agent has the right data to deliver the intended experience and drive outcomes for your audience is critical. In this section, you can configure the data sources your agent has access to.

Expand the 2. Agent Data Sources section in the top left of the screen. You should see the screen below. Here, you can take advantage of the existing data sources in watsonx Workshop or create your own.

There are two categories of data that you can provide to your AI Agent:

1. Static Knowledge​

Static Knowledge refers to information that you directly provide to your AI Agent, such as documents or webpages. Since this information does not automatically update after it is uploaded, it is considered static. However, you can return to your agent at any time to update or replace the information as needed.

When to Use Static Knowledge

  • Documents containing information that does not change frequently, such as sales frameworks or campaign materials.
  • Websites that contain information with infrequent changes, such as best practices for a specific process or activity.

Adding Static Knowledge

  • Under Add Files, upload one or more documents. The supported file formats are displayed on the screen.
  • Click Add under Add Web Content to specify a webpage using its URL. This can also include Seismic file URLs that you want to reference.

Note: Publisher pages are not currently supported, as they are not accessible within Workshop at this time.

2. Dynamic Knowledge​

Dynamic Knowledge refers to information that changes frequently. When connected to a dynamic data source, your AI Agent can retrieve information in real time during user interactions, ensuring access to the most up-to-date information without requiring manual updates to the agent itself.

When to Use Dynamic Knowledge

  • Your agent needs to search the web for current or real-time information.
  • Your content or datasets are updated regularly.
  • You want your agent to adapt based on changing business, industry, or client context.

Adding Dynamic Knowledge

  • To root your AI Agent in a specific IBM product, toggle Yes for Focus on Specific Product? and choose the relevant products from the dropdown menu. This allows your agent to provide detailed IBM product information sourced from IBM Docs and Seismic content.
  • To further tailor your AI Agent, you can root your session in a specific industry and sector by toggling Yes for Focus on Specific Industry? and select the relevant Industry and Sector(s).
  • Select Editable by users to control whether users can modify your configured product, industry, and sector selections. Enable this option to lock your configurations for all users.
  • Enabling Web Search allows your AI Agent to retrieve real-time information from the internet during user interactions.
  • Selecting Configure next to Enable Web Search allows you to customize how and when your AI Agent should search the internet. You can define when the agent should and should not search the web, as well as provide example web search behaviors to further refine how your AI Agent responds to user requests.

Step 3: Create the Agent Behavior

Your AI Agent’s behavior determines how it interacts with users, responds to prompts, and guides conversations throughout the experience. In this section, you can define your agent’s instructions, overall behavior, and greeting message to create the experience you want users to have.

Expand the 3. Agent Behavior section in the top left of the screen. You should now see the screen below.

There are two types of agent behaviors: Basic Agent is a lightweight agent powered by a single master prompt, while Skills Agent support multiple specialized skills, workflows, and tool-specific experiences for more advanced use cases.

Basic Agent

A Basic Agent is powered by a single master prompt that defines how the AI Agent behaves, responds to users, and guides conversations throughout the experience. Unlike Skills Agent, Basic Agent does not use specialized skills or workflows. To configure a Basic Agent, select the Basic Agent tab within the 3. Agent Behavior section. You can then define the following:

  • Agent Prompt: Define your AI Agent’s overall identity, purpose, scope, capabilities, behavior, and tone through a detailed system prompt. This prompt serves as the primary set of instructions that guides how your AI Agent responds to users.
  • Greeting Message: Customize the message users see when they begin a session with your AI Agent. This message should introduce the experience, set expectations, and guide users on how to get started. You can use the {firstName} variable to personalize the greeting message.

Skills Agent

A Skills Agent uses a master prompt combined with multiple specialized skills to deliver more advanced and task-specific experiences. Each skill can have its own behavior, tools, prompts, and workflows, allowing the AI Agent to support a wider range of use cases. To configure a Skills Agent, select the Skills Agent tab within the 3. Agent Behavior section. You can then configure the following:

  • Master Prompt: Define the foundational behavior, identity, purpose, scope, and tone that applies across all skills. This prompt acts as the backbone of the AI Agent, with individual skill prompts appended on top of it when a skill is engaged.

  • Skill Name: Define a short, user-facing name for the skill. This name is displayed to users and used by the router to identify the skill.

  • Description: Define when the skill should be used and the types of requests it is designed to support. This description is used by the router to determine when the skill should be engaged.

  • Stream from External Agent: Route the entire skill experience through an external service or agent instead of using a traditional Skill Prompt and Tools & Data configuration. When enabled, the external integration fully controls the skill behavior and response generation.

Use Stream from External Agent to stream all messages directly to or from an external agent that you specify. If you want to use an external agent as a sub-agent that provides context to this skill, where the agent still formulates responses to the user, leave this selection to Off and configure your agent as a datasource in the Tools & Data Sources below.

To configure:

  • Toggle Stream from External Agent to Yes.
  • Select Configure API.
  • Choose one of the following integration types:
    • External API to connect directly to any REST API endpoint using custom authentication, headers, request bodies, and variables.
    • A2A Agent to connect to an Agent-to-Agent (A2A) compatible external agent endpoint, commonly used with watsonx Orchestrate or other agent-based systems.
    • Proxy Agent to connect to an IBM Advisor Proxy streaming service with SSE support.
  • Skill Prompt: Define the behavior, instructions, scope, and tone specific to the skill. This prompt controls how the AI Agent behaves when the skill is activated.
  • Tools & Data: Configure the tools and data sources available to the skill. Depending on the use case, you can enable capabilities such as:
    • Product Context to provide access to the current IBM product and session context
    • Web Search to retrieve real-time external information
    • Start Presentation to generate presentations and slide decks
    • Start Roleplay to launch simulated sales roleplay conversations
    • External API to retrieve data from external API endpoints as additional context while still using the configured Skill Prompt and AI Agent behavior. When configuring External API under Tools & Data, the same integration configuration is used as Stream from External Agent. However, instead of replacing the skill behavior, the API acts as an additional context and data source available to the skill.
    • Start Podcast to generate podcasts from session content
    • Start Practice Gym to create interactive practice exercises and simulations
  • User Displayed Prompts: Define user-facing prompts that appear as clickable suggestions within the experience. Each displayed prompt can include an underlying prompt that is sent directly to the AI Agent when selected by the user.
  • Once all changes have been completed, select Save Skill Draft to save your skill configuration. You can add additional skills at any time by selecting Add Skill in the top right of the Skills section.

Step 4: Configure the Agent Studio Tools

The tools available within your AI Agent help create more interactive and engaging user experiences. In this section, you can configure the Learning Studio tools that users will have access to during their agent session, including podcasts, presentations, roleplays, and practice exercises.

Expand the 4. Agent Tools section in the top left of the screen. You should now see the screen below.

The Learning Studio Tools section allows you to configure the interactive experiences available to users during their AI Agent session. These tools can help users learn, practice, simulate conversations, and generate content directly within the experience.

Note: If you selected Skills Agent, ensure all skills have been fully configured before setting up the Learning Studio Tools.

Podcast

  • Toggle Podcast on to allow users to generate AI-powered podcasts from session content for on-the-go learning.
  • Select Set Default Prompt to configure the default instructions used when generating podcasts. Users can edit this prompt before generating their podcast.
  • Default podcast prompts should align to your AI Agent’s domain and intended learning outcomes.

Practice Gym

  • Toggle Practice Gym on to allow users to participate in quiz-based practice sessions that reinforce knowledge through interactive questioning.
  • Select Configure to define the topics users will be quizzed on during the experience. Aim for 3–5 focused topics relevant to your AI Agent.

Roleplay

  • Toggle Roleplay on to allow users to participate in simulated roleplay conversations or presentation-based practice exercises.
  • Select Configure to choose the roleplay experiences available to users. You can:
    • Select All Default to make all standard roleplay types available
    • Select Let Me Choose to customize roleplay scenarios tailored to your use case:
      • Under Default Roleplay Types, select the standard roleplay scenarios you want to make available, such as Cold Call or Discovery Call.
      • Under Your Custom Scenarios and Presentation Scenarios, select any existing custom roleplay experiences you want users to access.
      • Select Create new from Learning Studio to create a fully customized roleplay scenario tailored to your use case.
    • When creating a custom roleplay scenario, configure:
      • Select the Interaction type: Conversation for a 1-on-1 roleplay experience with an AI client or Presentation for a presentation-based practice session where users present independently.
      • Under Describe your practice scenario, provide detailed instructions describing the scenario, user objective, client context, AI persona behavior, and desired learning outcomes.
      • Once your scenario description is complete, select Next to continue configuring the roleplay experience.
    • Enable Lock configuration to prevent users from modifying the available roleplay types during their session.
    • Enable Allow users to create custom roleplay scenarios to allow users to build their own roleplay experiences in addition to the configured options.
    • Select Save once your roleplay configuration is complete.
  • Roleplay sessions are automatically graded based on the configured scenario and rubric.

Presentation

  • Toggle Presentation on to allow users to generate AI-powered presentations, including slide decks, transcripts, and AI-generated presenter videos.

Step 5: Test and Share Your Agent

Once your AI Agent has been created, use the Interactive Agent Test panel on the right side of the screen to validate the experience and ensure the agent behaves as intended. Test different prompts, workflows, and Learning Studio tools to refine your configuration.

If updates are needed, return to the relevant sections of the AI Agent Design flow and modify your configuration as required. Continue testing and refining your AI Agent until you are satisfied with the experience.

Once you select Finish, you will return to the My Agents page where your newly created AI Agent will appear in the table. From here, you can:

  • Select Update to continue making changes to your AI Agent configuration.
  • Select Delete to remove the AI Agent.
  • Select Share to generate and copy a shareable link to your AI Agent. After selecting Share, a modal will appear containing your AI Agent’s shareable link, which can be copied and shared across existing resources and collaboration channels, including presentations, Slack, Seismic pages, and other internal enablement materials.

Note: Currently, AI Agents can only be shared directly through their generated shareable link. Users with access to this link will be brought to the watsonx Workshop homepage and prompted to start a session with your AI Agent. They will not be able to edit or view the configured behaviors, prompts, tools, or data sources associated with the agent. Each AI Agent can only have one owner, meaning any future updates or configuration changes must be made by the original creator of the agent.