What is Agentforce
Agentforce is an enterprise-grade agentic AI platform from Salesforce. Announced on September 12, 2024 at Dreamforce 2024, expanded as Agentforce 2.0 in December 2024, and refreshed as Agentforce 360 at Dreamforce 2025 (October 14, 2025). It lets organizations build, test, deploy, monitor, and orchestrate autonomous AI agents grounded in Salesforce Customer 360 and Data Cloud data — across sales, service, marketing, commerce, and operations.
Architecture
At its core sits the Atlas Reasoning Engine — a reasoning runtime that interprets user intent, plans multi-step actions, and dynamically selects tools and APIs. The platform uses a BYO LLM model: by default it leverages partner models (OpenAI, Anthropic) and Salesforce's own models (e.g. xGen-Sales), configurable per agent. All LLM traffic passes through the Einstein Trust Layer, enforcing zero-retention with providers, sensitive-data masking, audit logging, and response toxicity detection.
Agent lifecycle
Agentforce 360 introduces the unified Agentforce Builder — a conversational workspace that merges design, test, and deploy phases. Three modes are available: low-code canvas, document-like editor with autocomplete, and a pro-code script view. The platform also ships Dev Tools (CLI, API, debugger), Agentforce Operations (monitoring, observability, cost), and AgentExchange (marketplace of prebuilt agents and actions).
Capabilities
Agents autonomously execute multi-step business workflows: qualify leads, answer service inquiries, draft sales emails, process returns, coordinate marketing campaigns, handle commerce transactions. Multi-agent routing, introduced in Agentforce 2.0, orchestrates multiple specialized agents collaborating within a single workflow. Agentforce Voice adds native voice channels (phone, voice channel).
Integrations & distribution
Agentforce inherits the full Salesforce integration footprint: Sales Cloud, Service Cloud, Marketing Cloud, Commerce Cloud, Data Cloud, MuleSoft, Tableau, Slack. It supports Model Context Protocol (MCP), so agents can call external tools and data sources via the MCP standard. Agents can be exposed in customer-facing channels (web chat, mobile, voice, Slack, WhatsApp) and internal-facing surfaces (Slack, Salesforce UI).
Pricing model
Salesforce charges through Flex Credits — credits are consumed proportionally to agent actions (early pricing: ~$0.10 per conversation). Agent features also ship as part of per-seat Sales/Service Cloud subscriptions and as enterprise add-on packages with dedicated SLAs and support.
