# Salesforce introduces architecture to manage AI agents > Source: > Published: 2026-09-10 14:19:25+00:00 **With Enterprise AI Harness, Salesforce is launching an overarching architecture to allow AI agents to function safely within companies. The platform combines six functional pillars with a central control panel, enabling organizations to manage and monitor both internal and external AI applications in a structured manner.** Salesforce introduces Enterprise AI Harness, a framework to allow AI agents to function in a controlled manner within business processes. Companies today struggle with the challenge that autonomous software agents can make decisions but often lack the broader business context. Without clear frameworks, agents may take actions that conflict with internal rules, compliance guidelines, or agreements made with customers. With the new architecture, the software company aims to bring data, business logic, and security together in one central location. As a result, IT departments do not have to reconfigure policies, permissions, and connections for each individual AI application. Customers’ existing systems, including Data 360, MuleSoft, and Agentforce, form the foundation for this expansion. ## Six pillars and central management The architecture rests on six fixed components encompassing context, action, agency, governance, security, and models. Trusted Context provides the necessary customer information and business data, while Trusted Action allows agents to actually perform tasks in external systems, such as adjusting an order via APIs. Trusted Security and data oversight guarantee that agents only access data they are authorized to see and that actions remain auditable. In addition, Salesforce is introducing a central control panel under the name AI Control Plane. This gives administrators a single overview of all active AI agents within the organization, regardless of whether they come from Salesforce itself or external providers. The control panel monitors performance, tracks operational costs, and checks whether the agents continue to operate within the established security guidelines. ## Open ecosystem and availability The platform is built according to an open structure and supports MCP. This allows companies to access functionalities directly from applications such as Microsoft Teams, Slack, or Claude. As a result, organizations are not tied to one specific interface or software, but retain the freedom to switch models as technology evolves. By focusing on reliability, Salesforce hopes to convince the final doubters of Agentforce. The AI platform around which Salesforce is building its ecosystem is [not yet meeting expectations.](https://itdaily.com/news/cloud/salesforce-agentforce-cant-convince/)