Agentforce is Salesforce's platform for building and deploying autonomous AI agents inside its CRM. Instead of a chatbot that answers questions, an Agentforce agent takes actions: it resolves service cases, qualifies inbound leads, books meetings, and updates records, grounded in a company's own Salesforce data. It launched in October 2024 and is priced on consumption, originally $2 per conversation and, since 2025, by the action through Flex Credits.
The architecture has a few moving parts worth knowing. The Atlas Reasoning Engine plans multi-step work rather than matching a question to a scripted reply. Each agent is configured from topics, which define what it is allowed to handle, and actions, which define what it can actually do: run a Flow, call Apex code, execute a prompt, or hit an external API. Data Cloud supplies the grounding, so answers draw on the company's records rather than the model's general knowledge. Agent Builder is the low-code surface where admins assemble all of this.
Salesforce ships prebuilt agents for the common jobs: a Service Agent for support, a Sales Development Representative for lead follow-up, a Sales Coach for rep training, plus agents for commerce and marketing tasks. Agentforce 3, released in June 2025, added a Command Center for observability and adopted Model Context Protocol support, which lets agents connect to external tools over an open standard rather than custom integrations.
The commercial model changed as the product matured. Launch pricing was $2 per resolved conversation. As of September 2026, pricing centers on Flex Credits: $500 buys a pack of 100,000 credits, a typical action draws 20 credits, and that works out to 10 cents per action. Consumption pricing tracks value more honestly than seats, but it makes budgeting harder, because a busy agent can burn credits faster than a finance team forecast.
Adoption is real but worth reading carefully. By mid-2025 Salesforce told investors it had signed more than 8,000 Agentforce deals since launch, roughly half of them paid. Salesforce also runs Agentforce on its own help site, which is the strongest evidence it offers: the company eats its own cooking at production scale. The open questions are the ones every enterprise agent platform faces: how often agents need human escalation, and whether per-action economics hold up once agents handle work in volume.
Go deeper: Salesforce's API-first agentic commerce push and how AI agents went corporate.