- Salesforce launched its Trusted Enterprise AI Harness on September 10, 2026, combining six governance capabilities with a new AI Control Plane.
- The full rollout of the Harness begins no earlier than February 2027, with two of the six announced capabilities not yet built.
- Salesforce’s Harness enters a market where Microsoft’s Agent 365 and AWS’s Amazon Bedrock AgentCore are already available.
Two of the six capabilities Salesforce announced in its new enterprise AI governance architecture don’t exist yet. The Trusted Enterprise AI Harness, unveiled on September 10, 2026, is a composable control layer for managing AI agents across heterogeneous stacks, but the full product won’t arrive until February 2027 at the earliest, leaving Salesforce playing catch-up in a market where Microsoft and AWS have already shipped competing control planes.
The company says enterprises are running an average of 3.1 agent orchestration platforms and that the majority use two or more concurrently, according to Salesforce’s own research.
The Harness is built around six capabilities: Trusted Context, Trusted Agency, Trusted Action, Trusted Governance, Trusted Security and Trusted Models. Trusted Context aggregates customer data, metadata, contracts, entitlements, real-time signals and memory into a single governed view, the idea being that an agent answering “Can we fulfill this order today?” needs to pull from CRM, ERP, analytics and policy simultaneously, not serially from separate systems. Trusted Agency handles multi-agent collaboration and complex reasoning over enterprise data. Trusted Action manages execution across enterprise systems, enforcing predefined business processes when an agent acts. Trusted Governance covers policy-setting, lifecycle management and compliance. Trusted Security protects sensitive data and constrains agent permissions. Trusted Models provides tooling for selecting, validating and monitoring underlying AI models, whether Salesforce’s own or third-party.
Salesforce was candid that two of those six capabilities are not yet fully built. The company did not specify which two.
The AI Control Plane sits above all six capabilities as a central management hub. It is designed to let enterprises discover and register agents, assign identity and policy, manage lifecycle, evaluate performance, observe behaviour and track costs, across both Salesforce and non-Salesforce agents. A July 2026 VentureBeat Intelligence survey of 107 respondents found that 53% of enterprises expect their primary agent control plane to be hybrid by the end of 2026, combining provider-native and external orchestration.
The competitive context matters here. Microsoft made Agent 365 generally available for commercial customers in May, positioning it as an enterprise control plane for observing, governing and securing agents beyond Microsoft’s own ecosystem. AWS introduced Amazon Bedrock AgentCore in June. Both are live. Salesforce’s Harness, by contrast, begins rollout in February 2027 at the earliest. The company’s Agent API, introduced in April 2026, had already signalled the intent to operate as a core execution layer for agentic AI, the Harness formalises that ambition into a full architecture. For more on how enterprise agent frameworks are currently stacking up the competitive picture is worth reading alongside this announcement.
Salesforce is assembling the Harness from existing platform components: Data 360, Informatica, MuleSoft and Agent Fabric, Tableau, Agentforce, Salesforce Guardian and the broader Salesforce Platform. MuleSoft Agent Fabric received a significant expansion in April 2026, adding automated discovery, centralised LLM governance and multi-vendor agent orchestration. That expansion now reads as the foundation the Harness is built on. The company’s Einstein Trust Layer, which includes secure data retrieval, data masking, zero retention policies, toxicity detection and audit trails, provides the security and compliance substrate. Salesforce’s published “Trusted AI Principles” (Responsibility, Accountability, Transparency, Empowerment, Inclusivity) are explicitly embedded in the Harness design, though how those principles translate into enforceable controls in practice will depend on capabilities that aren’t yet shipped.
Existing customers will be able to upgrade eligible Salesforce investments to access new capabilities as they roll out. Pricing, packaging and upgrade paths will be confirmed closer to general availability. What Salesforce hasn’t confirmed is a precise timeline for the two unfinished capabilities, or whether February 2027 is the start of a phased rollout or the date by which the full Harness is available. Given that Salesforce has been aggressive on agent announcements throughout 2026, the gap between architecture announcement and delivery is worth watching.
