started · updated
Salesforce introduces four-layer architecture for agentic AI
Salesforce has unveiled a new technological model focused on the transition from traditional software to the concept of the agentic enterprise. This model utilizes AI agents that work alongside humans by leveraging corporate data, processes, and systems under strict governance and control.
The proposed architecture consists of four distinct layers: data, applications, agents, and interface. The data layer provides the necessary context for agents to make precise decisions, while the application layer connects AI to business functions. Agents operate with defined roles and operational limits, and the interface serves as the primary contact point for human users.
A key component of this strategy is the integration of probabilistic reasoning from Large Language Models (LLMs) with the deterministic logic of CRM and business rules. This ensures that AI reasoning is combined with fixed business logic, permissions, and governance to achieve reliable execution at scale. Furthermore, the model allows for functionality to be accessed through external environments, such as Slack or Anthropic’s Claude, rather than being restricted to specific application interfaces.
Entities
Anthropic · Claude · Salesforce · Slack