Salesforce has introduced what it calls an Enterprise AI Harness, a common architecture designed to give AI agents access to business context, reasoning capabilities, actions and enterprise controls across different systems.
The architecture brings together six capabilities—Trusted Context, Trusted Agency, Trusted Action, Trusted Governance, Trusted Security and Trusted Models—alongside a new AI Control Plane that provides a central layer for managing AI agents and applications across an organisation.
Salesforce said the approach aims to tackle the challenges enterprises encounter as AI agents assume increasingly complex tasks across various functions and systems. to address the challenges enterprises face as AI agents take on more complex tasks across functions and systems. A customer query such as whether an order can be fulfilled, for example, may require information from CRM, ERP, contracts, analytics, policies, previous interactions and business processes. The AI Harness is designed to connect that information with the rules and controls required to execute the task.
The company said the architecture allows context, actions, workflows, governance and security controls to be shared across agents and models, rather than being developed separately for individual AI applications.
The six capabilities form the core of the architecture. Trusted Context combines customer data with metadata, semantics, knowledge, real-time signals and memory to give AI a common understanding of the customer and business. Trusted Agency provides reasoning, planning, state, memory, collaboration and orchestration capabilities, while Trusted Action connects AI to applications, APIs, workflows and business processes.
Trusted Governance covers data, metadata, policies and processes, including lineage, quality and guardrails. Trusted Security applies identity, permissions, privacy, data protection and runtime security to AI access and actions. Trusted Models allows enterprises to connect different AI models and route workloads based on factors such as accuracy, performance, cost and business requirements.
“The Agentic Enterprise won’t be defined by which model a company chooses,” said Rohan Kumar, president, chief platform and engineering officer, Salesforce. “What will differentiate an enterprise is the trusted, proprietary context it brings to that intelligence, starting with the customer, and its ability to securely turn that context into action.”
New AI Control Plane
Salesforce has also introduced an AI Control Plane to provide organisations with a central view of the AI operating in their environments. It is designed to help businesses discover and register agents and AI capabilities, establish identity and policies, manage lifecycles, evaluate performance, monitor behaviour and outcomes, and control costs across Salesforce and third-party AI systems.
The company said the control layer is intended to provide consistent visibility as AI agents and applications expand across teams, applications, models and systems, without requiring each agent to be managed independently.
The Enterprise AI Harness is being developed as a composable and open architecture. Salesforce said its capabilities will be accessible through technologies including MCP, APIs, skills and plug-ins, allowing them to extend beyond Salesforce applications into AI environments such as Claude, Slack and Microsoft Teams.
Shawn Malhotra, CTO, Rocket Mortgage, said the composability of the architecture was important as the AI landscape continues to change.
Salesforce said many of the technologies forming the foundation of the Enterprise AI Harness are already available, while new capabilities and the unified experience are expected to begin rolling out in early fiscal FY28. Existing customers will be able to upgrade eligible Salesforce investments as the new capabilities become available. Details on pricing, packaging and upgrade paths We will announce details on pricing, packaging, and upgrade paths closer to general availability.