Salesforce has introduced an enterprise architecture designed to give companies centralized control over artificial intelligence agents, models, data access, workflows, and security policies.
Called the Enterprise AI Harness, the architecture combines technologies from Salesforce’s Data 360, Informatica, MuleSoft, Agent Fabric, Tableau, Agentforce, Salesforce Guardian, and Salesforce Platform products.
The system is intended to provide AI agents with access to company data and business processes while limiting what information they can retrieve and what actions they are authorized to perform.
It consists of six capabilities covering context, agency, action, governance, security, and models. Companies may deploy them together or select individual components for use with Salesforce and third-party technologies.
The Trusted Context component combines customer records with metadata, business definitions, organizational knowledge, real-time signals, memory, and process information. Trusted Agency provides reasoning, planning, orchestration, and controls for tasks that require predictable outcomes.
Trusted Action connects agents to applications, application programming interfaces, workflows, and business processes. The governance and security components apply data lineage, quality controls, identity rules, permissions, privacy safeguards, and runtime protections.
Trusted Models, meanwhile, allows companies to connect different AI models and route tasks according to accuracy, performance, cost, and other business requirements.
“The Agentic Enterprise won’t be defined by which model a company chooses. Models will keep changing, and intelligence will increasingly be available everywhere. What will differentiate an enterprise is the trusted, proprietary context it brings to that intelligence — starting with the customer — and its ability to turn that context into action securely,” said Rohan Kumar, president and chief platform and engineering officer at Salesforce.
“That’s what we’re building with Salesforce’s Enterprise AI Harness: a system where every interaction and action can create new context and intelligence, making the enterprise’s context richer over time while operating with the security, governance, and control businesses need to trust AI at scale.”
Salesforce is also developing an AI Control Plane that would allow companies to register AI agents, assign identities and policies, manage their lifecycle, assess performance, monitor behavior and outcomes, and track costs.
The control plane will cover AI running on Salesforce as well as third-party systems, addressing the problem of agents being deployed separately across different teams, applications, and models.
Salesforce said the architecture is being developed with support for Model Context Protocol, APIs, skills, and plug-ins. This would allow its capabilities to operate through external services and applications such as Claude, Slack, Microsoft Teams, and Agentforce.
“AI is moving faster than any technology we’ve seen — which is exactly why composability matters so much to us. We don’t want to bet our future on one closed stack; we want the freedom to adapt as the landscape shifts. That’s what resonated with us about Salesforce’s new Enterprise AI Harness,” said Shawn Malhotra, chief technology officer at Rocket Mortgage.
Some of the underlying technologies are already available, but Salesforce said the new capabilities and unified interface would begin rolling out in its fiscal year 2028.
