Salesforce on Thursday unveiled its Salesforce Enterprise AI Harness, a platform designed to bring together fragmented artificial intelligence systems that currently operate in isolation across modern businesses. The company explained that no individual system holds all the information needed to handle basic business operations like fulfilling a customer order, where customer relationship management knows the client, enterprise resource planning tracks inventory, field service management handles delivery, and support teams manage processing. This fragmentation creates what Salesforce describes as AI leakage, where separate agents and their supporting infrastructure attempt to automate tasks within their own confined areas.
The Enterprise AI Harness brings together technologies spanning Data 360 for unified customer information, Informatica for data integration and oversight, MuleSoft and Agent Fabric for API connections and multi-agent coordination, Tableau for visual analytics, Agentforce as an agent platform, and Salesforce Guardian for security and compliance. These components work through a shared architecture and unified interface alongside a newly built AI control plane meant to support an open and composable AI ecosystem. The control plane offers businesses a centralized location to discover and register agents and AI functions, establish identity and policy frameworks, manage lifecycles, evaluate performance, observe behavior and results, and control expenses across both Salesforce and third-party AI systems. Developer feedback on existing Salesforce tools has pointed to challenges with initial setup and configuration, with one operations associate noting in August that building effective agents requires customization and workflow understanding, while a DevOps engineer commented in March that debugging and monitoring pipelines proves difficult in complex workflows and that the user interface needs improvement for better usability and faster navigation.
"The Agentic Enterprise won't be defined by which model a company chooses," said Rohan Kumar, Salesforce president and chief platform and engineering officer, during a press briefing. "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." According to Salesforce, the harness links reasoning to business rules, policies, and controls necessary for predictable execution, then makes those functions reusable across the organization so context can flow between agents and models. The company confirmed that software engineers can use all six capabilities together as a single system or select only what they need, creating deployments with Salesforce technology, other third-party existing technology, or both.
The platform rests on six core capabilities Salesforce labels as trusted functions. Trusted Context merges customer information with data, metadata, semantics, knowledge, real-time signals, memory, and an understanding of how work happens across the business. Trusted Agency provides agents with reasoning, planning, state, memory, and coordination abilities, blending flexible AI reasoning with deterministic controls where certainty matters. Trusted Action connects AI securely to applications, APIs, workflows, tools, and business processes, while Trusted Governance oversees the data, metadata, policies, and processes AI depends on with lineage, quality, guardrails, and controls. Trusted Security applies identity, permissions, privacy, data protection, and runtime security to what AI can access and what agents can execute, and Trusted Models ensures security through intelligent model routing based on accuracy, performance, cost, and requirements. The harness is being constructed headlessly from the ground up, with capabilities accessible through technologies including Model Context Protocol, APIs, skills, and plug-ins, allowing Salesforce functions to extend beyond traditional Salesforce applications into services like Claude, Slack, and Microsoft Teams.
Many foundational technologies are available now, with new capabilities and the unified experience scheduled to begin rolling out in early fiscal year 2028. The platform positions enterprises to move beyond model selection as the primary differentiator, instead emphasizing the proprietary context companies bring to intelligence and their capacity to convert that context into secure action. For businesses wrestling with siloed AI deployments and inconsistent governance, the challenge will be whether consolidation around a vendor ecosystem delivers meaningful efficiency gains or simply trades one form of complexity for another.