Assigning Tasks to AI Agents in Agentforce Operations

What AI Agents Are

You can assign tasks to AI agents and they perform focused business operations in your workflows. They process the task context and provided inputs, then return outputs based on their configured purpose.

When you assign a task to an agent, the agent analyzes the requirements and builds a plan to achieve the result. After the agent sets its plan, the agent follows it every time the task runs, which gives you predictable and consistent results.

Use agents for repeatable and rules-driven work, especially when the task result depends on structured data, attached documents, and clear completion criteria.

How to Use AI Agents

1. Select the right agent for the task. Define a clear objective and select the agent that matches it. For a list of available agents and their requirements, see AI Agents Catalog for Agentforce Operations.
2. Configure the task. Assign the agent and provide required fields, documents, and shared information from prior tasks. See Assign a Task to an AI Agent in Agentforce Operations.
3. Test the task and review the plan. Use the Test tab to run the agent and review the plan it generates. The plan defines the exact steps that the agent follows on every workflow run. See the How the AI Agents Plan Their Work section in Get AI Agent Tasks Ready for Run Time in Agentforce Operations.
4. Lock the plan by publishing the blueprint. After it's published, the agent follows the same plan every time. You can't publish a blueprint if an AI agent task is missing a saved plan. Tasks without a saved plan show a Missing plan badge on the task card. See the Locking an AI Agent's Plan section in Get AI Agent Tasks Ready for Run Time in Agentforce Operations.
5. Review output and handle exceptions. Check results in the task activity pane and route any exceptions to a human co-assignee or fallback assignee. See Add a Human Reviewer to an AI Agent Task in Agentforce Operations.

How AI Agent Plans Work

When you assign a task to an AI agent, it analyzes the task configuration and creates a plan before the workflow runs. The plan defines the exact steps that the agent follows every time, ensuring consistent results across workflow runs.

Agentforce Operations creates plans during blueprint development and locks them when you publish the blueprint. After the plan locks, the agent follows the same plan for every workflow instance, which prevents variability and ensures repeatable outcomes. If the task's configuration changes, regenerate the plan by testing the agent again.

You can't publish a blueprint until all AI agent tasks have saved plans. Tasks without plans show a Missing plan badge on the task card. To learn more, see the Locking an AI Agent's Plan section in Get AI Agent Tasks Ready for Run Time in Agentforce Operations.

Testing AI Agents

Test AI agent tasks before publishing to verify that the agent produces the expected output. Use the Test tab in the task configuration panel to run the agent with sample input data and review the generated plan. Testing helps you catch configuration issues early and makes sure that the agent understands the task requirements.

During testing, provide realistic input examples that represent the variety of data the agent can encounter at run time. Review both the plan steps and the output values to confirm they match your expectations. To learn more, see the Testing AI Agent Tasks section in Get AI Agent Tasks Ready for Run Time in Agentforce Operations.

Agent At Run Time Behavior

After the task starts, the agent begins work immediately. Most tasks complete in under a minute. Use the task activity pane to review the full history.

If the agent encounters an error, there are several different paths the agent can take.

• If the task doesn't have co-assignees or fallback assignees, the agent reports a problem on the task.
• If the task has a human co-assignee, the agent doesn't complete the task. The human assignee takes over.
• If the task has fallback assignees configured, the task gets reassigned to the fallback users.

Working with AI Agents

Review the limits and considerations for AI agent tasks, including supported file types, field limits, and timing constraints. To learn more, see AI Agent Limits and Considerations in Agentforce Operations.

Write clear, specific instructions that help the agent produce accurate results. Good prompts define the goal, expected output format, and any constraints. To learn more, see Write Effective Instructions for AI Agent in Agentforce Operations.

Diagnose and resolve issues when an AI agent task doesn't complete as expected. To learn more, see Troubleshoot AI Agent Tasks in Agentforce Operations.

To learn more, explore these links.