Agent creation specifications

The agent creation specifications provide you a high-level understanding of the fundamental components that form an AI agent. Understanding the details can help you streamline and simplify the agent design process and provide a user-friendly experience.

Components of an agent

By understanding the components of an agent, you can create more effective and efficient agents tailored to specific use cases and requirements. The following components act a blueprint for designing and developing agents:

  • Profile: Represents the agent's identity, including its name, role, and any other relevant attributes. This information helps define your agent's purpose and its context.

  • Knowledge: Encapsulates the information and data your agent has access to, which it uses to make decisions and perform tasks.

  • Toolset: Refers to the tools that your agent employs to process information, make decisions, and interact.

  • Behavior: Describes your agent's actions and responses, based on its profile, knowledge, and toolset. It outlines how your agent interacts with users, other agents to achieve its goals.

  • Channels: Connects your agent to the communication platforms (for example, Slack). This integration improves user experience and agent accessibility.

Orchestration of agents

  • In watsonx Orchestrate, your agent can collaborate with other agents and tools to achieve common goals or complete tasks. Agent collaboration enables improved performance, scalability, and adaptability compared to single-agent systems.

  • watsonx Orchestrate uses description-based routing to call or reference other agents added as collaborators. When a user interacts with an agent, the Orchestrator examines the agent's tools and collaborators, and based on the descriptions of these artifacts, it determines the optimal routing for the interaction. By adding a detailed description to your agent, you help watsonx Orchestrate efficiently manage and coordinate the collaboration between agents. See Recommendations for agent descriptions for more details.

  • The collaborator agents on watsonx Orchestrate are run sequentially. The tasks are completed in a specific order, ensuring that dependencies are respected and resources are used efficiently.

  • There are no strict restrictions on the number of collaborator agents or tools, but a larger toolset may require more accurate descriptions for optimal routing. It is recommended that you keep the toolset small to ensure the best routing results and improve the overall performance of your multi-agent system.

Change handling

  • Changes that you make on an agent take effect in the live environment only after the agent is deployed. Any modifications that you make to an agent in the draft environment will not be visible in the live environment until your agent is deployed.

  • When you edit an agent, the changes are immediately reflected in the draft environment, allowing you to test and verify the updates before deploying. However, these changes do not propagate to other agents using it as a collaborator until the edited agent is deployed in the live environment.

  • Editing one agent does not automatically update other agents using it as a collaborator; collaborator agents will only see the changes once the edited agent is deployed.

Design implications

Performance considerations: Having a larger toolset can indeed have performance impacts on your agents, primarily due to increased complexity and longer routing times. It's essential that you optimize tool selection and design efficiently to streamline processes.

Agent validation: Test your agents individually before adding them as collaborator agents of other agents to avoid unexpected outputs.