Salesforce Launches Enterprise AI Harness and AI Control Plane Ahead of Dreamforce

Quick Facts

Salesforce introduced the Trusted Enterprise AI Harness on Sept. 10, built around six capabilities including Trusted Context, Trusted Security, and Trusted Models.
A new AI Control Plane lets companies discover, register, monitor, and manage AI agents from both Salesforce and third-party providers.
Gartner predicts 40% of enterprise applications will integrate task-specific AI agents by year-end 2026, up from less than 5% in 2025.

Salesforce unveiled the Trusted Enterprise AI Harness on Sept. 10, days before Dreamforce 2026, which runs Sept. 15-17 at the Moscone Convention Center in San Francisco. The company is centering the event on what it calls the Agentic Enterprise.

The Harness is a composable architecture that bundles six capabilities: Trusted Context, Trusted Agency, Trusted Action, Trusted Governance, Trusted Security, and Trusted Models. Companies can adopt all six as one system or pull in only what they need, alongside their existing tools and third-party models.

What the Harness Does

When developers build an AI agent, they extend a large language model with add-ons such as prompts, database connectors, and other technical assets. Salesforce calls that collection of assets a harness. By placing those capabilities in a shared harness rather than inside individual agents, Salesforce says enterprises can reuse context, actions, and governance controls across every agent without rebuilding them separately each time.

The Harness draws on Data 360, Informatica, MuleSoft, Agent Fabric, Tableau, Agentforce, Salesforce Guardian, and the Salesforce Platform. It is built headlessly, with capabilities reachable through MCP, APIs, Skills, and plug-ins, extending to Claude, Slack, and Microsoft Teams.

Rohan Kumar, Salesforce President and Chief Platform and Engineering Officer, said the differentiator for enterprises will not be the model itself. “Models will continue to change, and intelligence will increasingly be available everywhere,” Kumar said. “What will differentiate an enterprise is the trusted, proprietary context it brings to that intelligence.”

The AI Control Plane

Alongside the Harness, Salesforce introduced an AI Control Plane that gives organizations a single place to see, manage, and control agents and AI across the business. The Control Plane covers agent discovery and registration, identity and policy, lifecycle management, performance evaluation, behavioral monitoring, and cost control.

Kumar described the Control Plane as an evolution of the API management tooling Salesforce already sells through MuleSoft, extended to cover MCP servers, LLM endpoints, and an agent registry. It functions similarly to ServiceNow’s AI Control Tower, Microsoft’s Agent 365, Amazon Bedrock AgentCore, and SAP’s AI Agent Hub.

Shawn Malhotra, CTO of Rocket Mortgage, cited composability as a central concern. “We don’t want to bet our future on one closed stack; we want the freedom to adapt as the landscape shifts,” Malhotra said.

Market Pressure Behind the Move

The timing reflects real pressure on enterprise software vendors. A July 2026 VentureBeat Intelligence survey of 107 organizations with 100 or more employees found that 85% run two or more agent orchestration platforms simultaneously, with an average of 3.1 platforms per enterprise. Another 53% expect their primary agent control plane by year-end to be hybrid, combining provider-native and external orchestration.

Gartner’s forecast makes the business case clear: 40% of enterprise applications will carry task-specific AI agents by the end of 2026, compared with less than 5% in 2025. That is an eightfold increase in a single year, and it signals the window for platform vendors to establish control-plane dominance is short.

Salesforce also announced Slack Code and Claudeforce ahead of Dreamforce, signaling a broader push to position its platform as the connective layer for AI work across the enterprise.

This article was written by an AI agent. Spotted an error? Send a correction and we will fix it.