Understanding Agentforce Agents

Agentforce agents are primarily ReAct agentic loops, which means they use LLMs to understand utterances and reason about topics. Once a topic is selected, the agent uses an LLM to plan tool calls, makes the tool calls, and then reasons about the output to decide whether to loop or not.

Further reading on the topic can be found on the Salesforce website, which provides more information on ReAct agents and the reasoning engine behind Agentforce.

Building Functionality with Agentforce Agents

When using an Agentforce agent, teams can build functionality using Apex, LWC, or Flows, and then integrate it with the agent. The agent’s behavior depends on the defined topic, and the reasoning instructions that tell the agent how and when to use the tools.

Agent Action Example

public class MyAgentAction {
    public String execute() {
        // Tool call logic here
        return 'Tool call result';
    }
}

The overall behavior of the agent is a mixture of deterministic and LLM reasoning, with the actual capabilities for action defined by the tools provided.

The root cause of the intelligence behind Agentforce agents is the combination of LLMs and system prompts, which enable the agent to understand natural language and reason about topics.

Best Practices for Using Agentforce Agents

While Agentforce agents can be powerful tools, they may not be the correct answer for every use case. However, by expanding the capabilities of the agent, teams can achieve more complex tasks, such as querying and finding relationships between records.

Heads up: Agentforce agents are not a replacement for traditional coding, but rather a complementary tool that can enhance the capabilities of your team.

Checklist for Using Agentforce Agents

• Define clear topics and reasoning instructions for the agent
• Build functionality using Apex, LWC, or Flows
• Integrate the agent with your existing tools and systems
• Test and refine the agent’s behavior
• Continuously monitor and evaluate the agent’s performance

What is the primary function of an Agentforce agent?

The primary function of an Agentforce agent is to understand natural language and reason about topics, using a combination of LLMs and system prompts.

Can I use Agentforce agents for every use case?

No, Agentforce agents may not be the correct answer for every use case. Teams should evaluate the capabilities and limitations of the agent before deciding to use it.

How do I build functionality with an Agentforce agent?

Teams can build functionality using Apex, LWC, or Flows, and then integrate it with the agent.

What is the difference between deterministic and LLM reasoning?

Deterministic reasoning uses predefined rules and logic, while LLM reasoning uses natural language processing and machine learning algorithms to understand and respond to queries.