NitiExam
CRM & SaaSv2.0Updated 2026-07

AgentForce

AgentForce is Salesforce's AI agent platform for building autonomous agents that perform CRM tasks without human intervention — answering customer queries, updating records, escalating cases, and processing orders.

Overview

  • 1AgentForce (GA: Oct 2024, v2.0: Feb 2025) lets Salesforce admins build AI agents in Agent Studio.
  • 2Agents are composed of Topics (goals), Actions (tools), Instructions (behaviour), and Guardrails.
  • 3AgentForce uses Einstein AI reasoning (built on LLMs) to dynamically select actions per query.
  • 4Pre-built agents: Salesforce SDR Agent, Sales Coach Agent, Service Agent, Merchant Agent.
  • 5Agents can handle multi-turn conversations and take actions across Salesforce and external systems.

Key Features

Agent Studio — no-code visual builder for creating and testing agents
Agent Testing Center — simulate conversations and validate agent responses
Topics — goal-based instructions that activate when user intent matches
Actions — Apex, Flow, prompt templates, or external API calls that agents can invoke
Guardrails — prohibited topics and response length constraints
Data Cloud grounding — agents query unified customer data for personalised responses
Omni-channel deployment — agents work in messaging, email, web chat, and phone

How AgentForce Works

  • User sends a message → Einstein Reasoning selects relevant Topic from the agent
  • Einstein maps user intent to an Action defined on the Topic
  • Action executes (Flow, Apex, prompt template, HTTP callout) and returns a result
  • Agent synthesises result into a natural language response and optionally takes follow-up steps
  • Handoff: agent transfers to human agent if Guardrail confidence threshold is not met
  • All interactions are logged in the Conversation record for analytics and audit

Building an Agent

  • Setup → AgentForce → Create Agent: choose from template or start blank
  • Define Topics: "Order Status" topic activated when customer asks about their order
  • Add Actions to Topics: "Get Order Record" Action queries Order object via SOQL
  • Write Instructions: tell the agent how to respond ("Always confirm order number before looking up")
  • Set Guardrails: "Never discuss competitor pricing", max message length: 500 chars
  • Test in Agent Testing Center: submit test queries and review reasoning trace
  • Deploy: embed in Experience Cloud site, Service Console, or external messaging via API

AgentForce Actions

  • Apex Action: call an @InvocableMethod Apex class — most powerful, fully customisable
  • Flow Action: invoke a Screen or Auto-launched Flow — low-code, reusable
  • Prompt Template Action: generate text via Einstein LLM with grounded Salesforce data
  • API Action: call external REST endpoint via Named Credential — no credential exposure
  • Standard Actions: built-in actions for Case Create, Knowledge Search, Contact Lookup
  • Data Cloud Action: query unified customer profiles and predicted scores from Data Cloud

Agent Analytics & Optimisation

  • AgentForce Analytics: resolution rate, deflection rate, CSAT correlation, handoff triggers
  • Conversation Inspector: drill into individual agent runs — reasoning trace, actions invoked
  • A/B Testing: compare Topic instruction variants with statistical significance
  • Feedback loop: thumbs up/down ratings from agents train Einstein ranking improvements
  • Latency monitoring: P95 response time per action; identify slow Apex or external callouts

Frequently Asked Questions

What is AgentForce and how is it different from Einstein Bots?

AgentForce is a full AI agent platform powered by large language models and Einstein reasoning. Einstein Bots (the legacy chatbot tool) use decision-tree dialog flows with predefined intents. AgentForce agents understand natural language, reason about which actions to take, and can perform multi-step tasks autonomously without scripted dialog paths.

Do I need to code to build an AgentForce agent?

No. Agent Studio provides a no-code interface for creating Topics, adding Actions (including built-in standard actions and existing Flows), writing Instructions, and testing the agent. However, to build custom Actions that access specific data or external systems, you will need Apex (@InvocableMethod) or a declarative Flow — which require Salesforce developer skills.

How does AgentForce handle sensitive customer data?

AgentForce inherits Salesforce's object-level and field-level security. Agents only access records and fields that the agent's running user profile permits. Guardrails can prohibit the agent from discussing or displaying specific fields. All agent conversations are stored in Salesforce and subject to the org's data residency, encryption, and retention policies.