What it is

Agentforce is Salesforce’s platform for building AI agents that act inside the Salesforce ecosystem — sales, service, marketing, commerce. You build agents with a low-code builder and they can take real actions (update records, send emails, open cases) inside the CRM.

Best for

Service agents that handle tier-1 support cases in Salesforce
Sales agents that qualify and route leads
Marketing agents that personalize outreach

Pros

Native to Salesforce — no integration tax
Low-code builder for non-developers
Enterprise-grade security and compliance

Cons

Locked into the Salesforce ecosystem
Per-conversation pricing adds up

Quick start

Sign in to Salesforce → App Launcher → "Agentforce Builder" (added to your org with Einstein AI enabled).
Einstein AI + 1 Agentforce user license = roughly $2 per conversation billed at runtime.
Open the "Agent Builder" canvas; pick a template (Sales Coach, Service Agent, SDR, Personal Shopper) or start blank.
Drop in "Topics" (what the agent handles) → Actions (Flows + Apex invocable methods + HTTP calls).
Connect Data Cloud / Knowledge articles so the agent pulls the right context and grounds its responses.
Test in the "Agentforce Testing Center" with synthetic conversations, then deploy to a Service Cloud channel (chat, SMS, voice).

Sample input / output

Input

Topic: "Order Status" (Service)
Definition:
  Trigger phrases: "where's my order", "tracking number", "check delivery"
  Instructions:
    - Always ask for order number (format: ORD-XXXX).
    - Pull order from Commerce Cloud via getOrderStatusById action.
    - If ETA is within 2 business days, do NOT mention delays.
    - For damaged-on-arrival, open a "Damaged Shipment" case via Flow.

Action: getOrderStatusById
  Input: { orderId: string }
  Output: { eta: date, courier: string, status: enum }

Channel: Service Cloud Web Chat (production)
Escalation rule: if customer sends "supervisor" or "manager" → route to human queue "Tier-2 NA"

Output

Day-1 rollout (Service Cloud, NA queue):
  Conversations routed: 2,310
  Resolved without human: 1,820 (78.8%)
  Avg handle time: 1m 42s (was 6m 12s with full human)
  CSAT: 4.51 / 5
  Escalations to Tier-2: 290 (12.5%)
  Per-conversation cost: $1.95 average
  Hidden catch: 8% of conversations triggered Flow regressions from a Salesforce Spring '26 release — re-ground your Topics when the next release ships.

Benchmarks

Launched Dreamforce 2024; GA since Feb 2025 — Salesforce

Per-conversation pricing From $2.00 / conversation — Salesforce pricing 2026

Slack-native Build agents inside Slack channels since 2025 — Salesforce+Slack

Atlas Reasoning Engine New reasoning layer shipped Feb 2026 — Salesforce Dreamforce 2025

Einstein Trust Layer Zero data retention by default; Data Cloud grounding — Salesforce security brief

Pricing

From $2/conversation

Underlying models

Einstein / OpenAI / Anthropic (via Salesforce)

Should you pick this?

Pick it if Your revenue, service, and contact-center workflows already live in Salesforce. Agentforce eliminates integration cost and inherits your existing security model.

Skip it if You want a CX platform built for non-Salesforce stacks (Sierra is stack-agnostic), or you want pricing per resolved ticket with analytics out of the box (Decagon). For pure voice automation at scale, Vapi or Retell.

Salesforce Agentforce