Section 4 · AI Agents · 35% of exam
4.7 Standard and Custom Agent Actions
Exam objective
Given a use case, choose a standard or custom Agentforce action and understand how its instructions, inputs, and outputs affect execution.
Actions are the agent’s tools
An action gives the agent a capability it can invoke, such as finding information, creating or updating a record, running a Flow, calling Apex, or using a prompt template. The agent reasons about whether an available action fits the request, based on its instructions, inputs, outputs, and the context of the interaction. An action is not just a response paragraph: it can perform work in Salesforce or another supported system. Keep the action’s business purpose clear and make sure its behavior matches the authority granted to the agent.
When to use standard actions
Standard agent actions are provided by Salesforce and may support common tasks or specific products. Many are available to users with permission to use Agentforce, while some require an additional license or permission. A standard action can still need configuration. Salesforce directs admins to customize the underlying Flow, prompt template, or reference action for the intended use case. Review the action description, required inputs, returned outputs, limits, and product availability before adding it. Do not assume a standard action is enabled, licensed, or appropriate for every agent.
When to build a custom action
A custom action is appropriate when the agent must perform a business operation that the available standard actions do not cover in a suitable way. The action can expose an existing automation or service capability to the agent, with defined inputs and outputs. Choose a supported implementation that fits the task and the organization’s development standards. Keep inputs narrowly scoped and understandable; do not make the model supply values that can be obtained more reliably from trusted context or a prior action. Output descriptions should state what the result means so the agent can use it correctly. Apply least privilege and validate data at the underlying action layer. Make actions predictable
An action’s availability does not ensure it will be selected at the right time. Provide a concise description of when it should and should not be used. Instructions should clarify prerequisites, required information, and what to do with the result. If an operation requires exact sequencing or a hard condition, control it with Agent Script, filters, Flow, or another deterministic mechanism rather than relying only on the model. Handle errors and empty results explicitly. Test correct input, missing input, invalid values, permission denial, and service failure. Verify the actual record change or output; do not judge success solely from the agent’s natural-language confirmation.
TIP: Match the action to the operation: use a standard action when its behavior and licensing fit; configure or create a custom action when the use case needs different logic.
Action instructions influence the agent’s choice and use of the action, while the underlying Flow, Apex, or other reference implementation performs the operation.
Extra concept: An action output can be mapped to a variable and reused in instructions, filters, or a later action, making multi-step tasks more controlled.
Official references:
- Standard Agent Action Reference https://help.salesforce.com/s/articleView?id=ai.agent_actions_ref_pointer.htm&language=en_US&type=5
- Configure Your Agent https://help.salesforce.com/s/articleView?id=ai.agent_parent_configure.htm&language=en_US&type=5
- Create a Custom Agent Action https://help.salesforce.com/s/articleView?id=xcloud.archive_create_a_custom_agent_action.htm&language=en_US&type=5
Key takeaways
- Actions provide operational capabilities, while instructions help the agent select and use them.
- Check standard action requirements and customize its underlying implementation for the use case.
- Use deterministic controls and test failure paths for high-impact operations.
