Salesforce Agentforce

Salesforce launched Agentforce at Dreamforce in September 2024 and made it generally available in October 2024, positioning it as the third wave of AI after predictive AI and copilots. Unlike copilots that assist humans, Agentforce deploys agents that act autonomously within defined guardrails.

The platform is built on the Atlas Reasoning Engine, Salesforce's proprietary reasoning layer that combines planning, action selection, and execution. Agents have access to 200+ pre-built actions covering CRM operations, data lookups, case management, lead qualification, appointment scheduling, and order management. They can also call external APIs and trigger flows.

Agentforce agents are configured rather than coded. Administrators define an agent's role, the topics it can handle, available actions, and guardrails (when to escalate to a human). The builder interface uses natural language to set instructions. No machine learning training or data labeling is required — agents learn what they can do from the action library and the instructions you give.

The primary deployment channels are Service Cloud (customer-facing chat for support deflection), Sales Cloud (SDR tasks like lead follow-up and meeting scheduling), and Marketing Cloud (campaign response and lead nurturing). Agentforce is also available as a component in Slack and email.

Pricing is $2 per conversation for Agentforce Service Agent and Agentforce Sales Development Representative. Enterprise agreements with volume pricing are available. Agentforce is included in certain Salesforce bundles at higher tiers.

As of early 2026, Salesforce reported 5,000+ Agentforce customer deployments and claims ROI through call deflection, faster lead response, and reduced agent staffing costs.