Identify the Subagents and Actions

When identifying the subagents and actions for your agent, there are two possible approaches: top-down and bottom-up. In the top-down approach, first identify the subagents, then define the relevant actions for those subagents. In the bottom-up approach, first list all of the individual actions for your agent, then group them into related subagents.

Bottom-up Example:

List the Actions

To come up with ideas for agent actions, review the business processes related to the work your AI agent will do, and then list all the tasks the agent should be able to perform. At this stage, it's better to be too specific rather than too general. Avoid trying to group or simplify actions prematurely. The goal is to create a comprehensive and granular view of what the agent can do.

For example, in the case of a customer service agent, the initial list might include:

• Answer general FAQs
• Find customer orders
• Look up items related to an order
• Check order status
• Check delivery status
• Return an order
• Exchange an order
• Look up return and exchange policies
• Check inventory availability
• Check promotion availability
• Troubleshoot technical problems
• Find customer assets and warranties

Note that an action like “Resolve customer complaints” is too broad because customers can have different types of complaints: problems with an order, technical issues, or delivery questions. Actions should represent the smallest level of granularity in agent behavior.

Identify Overlapping or Similar Actions

Flag actions that are too similar because they can cause confusion for the reasoning engine. If the actions aren’t different enough semantically, then the reasoning engine won’t know which action to select when a user asks a question or makes a request.

For example, “Troubleshoot technical problems” and “Answer general FAQs” might have descriptions that are similar, but their functionality is different. Identifying semantic overlaps can help to determine which actions to separate over multiple subagents.

Group Actions into Subagents

After actions are clearly defined, they can be grouped into preliminary subagents. A subagent is a logical category of functionality—a grouping of actions that together represent a job, capability, or skill of the agent.

Follow these guidelines:

• Aim to use the minimum number of subagents necessary to avoid semantic overlap.
• Make sure that each subagent contains actions that are meaningfully related.
• If certain actions must be executed in a chain, include them in the same subagent.

Here’s an example of the subagent and action groupings for a customer service agent:

Subagent 1: Order Management

• Find customer orders
• Look up items related to an order
• Check order status
• Check delivery status
• Check inventory availability

Subagent 2: Troubleshooting

• Troubleshoot technical problems
• Find customer assets and warranties

Subagent 3: Returns and Exchanges

• Look up return and exchange policies
• Return an order
• Exchange an order

Subagent 4: Product Support

• Answer general FAQs
• Check promotion availability

Tip You can build your Agentforce subagents and actions from scratch, but there are also templates you can use to get up and running quickly. Another fast way to build subagents and actions is to create them using generative AI.