Salesforce on September 10, 2026 introduced the Trusted Enterprise AI Harness, a composable architecture built around six trusted capabilities and a new AI Control Plane that the company says gives AI a shared understanding of the customer and the business while keeping its actions within enterprise controls.
The announcement frames the harness as an answer to a specific problem of the Agentic Enterprise. As agents take on more complex work, understanding what is happening, deciding what to do next, taking action across systems, and working alongside people and other agents, Salesforce says the challenge becomes equipping them to do that work reliably, securely, and at scale without building and managing those capabilities separately for every agent or AI experience.
Salesforce illustrates the gap with a single customer question: “Can we fulfill this order today?” In the company’s telling, no single system holds the complete answer. CRM knows the customer and the relationship, ERP knows inventory and fulfillment, contracts contain commitments and entitlements, analytics provides business definitions, policies determine what can be promised, previous interactions supply memory, and business processes determine how the work should get done. The challenge goes beyond connecting those systems: an agent must understand all of that information in the context of a specific customer at a specific moment, then determine the right action, take it securely, and operate within the policies and controls of the business. The announcement notes that AI reasoning can be open-ended while enterprise execution often cannot, so the harness connects that reasoning to the business rules, policies, and controls required for predictable execution.
Because the capabilities live in a shared harness rather than inside individual agents, Salesforce says they become reusable across the enterprise: context can be shared across agents and models, actions and workflows can be securely invoked wherever they are needed, and governance and security can be applied consistently as AI moves across the business. The announcement also describes a compounding effect in which every interaction and action creates new signals, outcomes, and memory that enrich the context available to what comes next. Models will continue to improve and become more broadly available, the company argues, while an enterprise’s proprietary context, spanning its customers, knowledge, semantics, memory, processes, relationships, and history, remains uniquely its own and compounds into what Salesforce calls a durable source of differentiation.
Salesforce says it is building the harness on core technologies from across Data 360, Informatica, MuleSoft and Agent Fabric, Tableau, Agentforce, Salesforce Guardian, and the Salesforce Platform, delivered through a common, composable architecture and unified experience, and on the customer relationships, processes, and controls already running the business.
The Six Trusted Capabilities
According to Salesforce, trust is designed into each of the six capabilities from the start, and each plays a different role in answering the order question.
Trusted Context brings together customer context with data, metadata, semantics, knowledge, real-time signals, memory, and an understanding of how work gets done across the enterprise. Grounded in governed enterprise data, it is meant to give AI a shared understanding of the customer and the business around them; for the order, that means understanding the customer, their contract and entitlements, available inventory, and what “available” means for that customer.
Trusted Agency supplies the reasoning, planning, state, memory, collaboration, and orchestration agents need to pursue complex business outcomes, combining flexible AI reasoning with deterministic controls where certainty is required. In the example, it determines whether the order can be fulfilled and what needs to happen next while following the business rules that require certainty.
Trusted Action securely connects AI to applications, APIs, workflows, tools, and business processes, allowing an agent to reserve inventory, update the order, trigger fulfillment, or engage a person when needed, with the outcomes of those actions flowing back into Trusted Context.
Trusted Governance covers the data, metadata, policies, and processes AI relies on, with lineage, quality, guardrails, and controls intended to keep answers based on trusted information and the agent within the policies governing how the order should be handled.
Trusted Security applies identity, permissions, privacy, data protection, and runtime security to what AI can access and what agents are allowed to do, so an agent sees only the customer information and takes only the actions it is authorized to access.
Trusted Models lets enterprises securely connect the right model for each job, with intelligent routing based on accuracy, performance, cost, and business requirements, while giving the business flexibility to change models as the technology evolves.
“The Agentic Enterprise won’t be defined by which model a company chooses,” said Rohan Kumar, Salesforce’s President, Chief Platform and Engineering Officer. He said 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. Kumar described the harness as a system in which 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.
A New AI Control Plane for the Enterprise
Alongside the six capabilities, Salesforce is introducing a new AI Control Plane that gives businesses one place to see, manage, and control agents and AI across the enterprise. The company says the Control Plane enables companies to discover and register agents and AI capabilities, establish identity and policy, manage lifecycle, evaluate performance, observe behavior and outcomes, and control cost, across both Salesforce and third-party AI. Salesforce frames it as a consistent layer of visibility and control as AI expands across teams, applications, models, and systems, without requiring every agent or AI experience to be managed separately.
Composable, Headless, and Open to Third Parties
Customers can use the six capabilities together as one system or take only what they need, with Salesforce technology, their existing technology, or both, including third-party models, agents, and systems. The harness is being built headlessly from the ground up, with capabilities accessible through technologies including MCP, APIs, Skills, and Plug-ins, which Salesforce says allows them to extend beyond traditional Salesforce applications and into the AI experiences where people already work, including Claude, Slack, Microsoft Teams, and Agentforce. The company says the design advances its broader AIforce strategy to bring Salesforce to any AI, agent, app, or surface.
A new unified experience is planned so that different teams work with the capabilities most relevant to them: context and governance for data leaders, identity and policy for security leaders, and agents, models, and actions for builders.
Rocket Mortgage CTO Shawn Malhotra said in the announcement that AI is moving faster than any technology his company has seen, which is exactly why composability matters to Rocket Mortgage: it wants the freedom to adapt as the landscape shifts rather than bet its future on one closed stack. That, he said, is what resonated with the company about Salesforce’s new Enterprise AI Harness.
Availability
Salesforce says many of the technologies that form the foundation of the Trusted Enterprise AI Harness are already available, with new capabilities and the unified experience planned to begin rolling out in early fiscal FY28. Existing customers will be able to upgrade eligible Salesforce investments to unlock new capabilities as they become available. Additional details on availability, packaging, pricing, and upgrade paths will be announced closer to general availability, the company said.
Credit: Source link
