Salesforce unveils Enterprise AI Harness to govern AI agents across multi-platform environments

Salesforce Tackles Multi-Platform AI Agent Reality

Salesforce previewed its Trusted Enterprise AI Harness ahead of the Dreamforce conference next week in San Francisco, targeting a critical challenge facing enterprises today. VentureBeat Intelligence's July 2026 Agentic Orchestration Pulse Survey reveals that 85% of organizations run two or more agent orchestration platforms simultaneously, averaging 3.1 platforms per enterprise1

1

2

2

. Additionally, 53% expect their primary agent control system by the end of 2026 to be hybrid, combining provider-native and external orchestration1

1

.

The Enterprise AI Harness bundles six capabilities—Trusted Context, Trusted Agency, Trusted Action, Trusted Governance, Trusted Security, and Trusted Models—alongside a new AI Control Plane for managing agents and AI across organizations3

3

. While many technologies underpinning the offering are already available in Salesforce's cloud services, full availability won't arrive until early in the company's 2028 fiscal year, beginning next February3

3

.

Proprietary Business Context as Competitive Advantage

Rohan Kumar, Salesforce's president and chief platform and engineering officer, argues that as foundation models become more widely available, durable enterprise advantage will stem from proprietary business context and operational controls placed around them1

1

. "The intelligence that [is] coming from the model is ubiquitous," Kumar said. What remains unique is an organization's "enterprise trusted context": its customers, employees, transactions, data relationships, knowledge and history2

2

.

Secure AI agents need more than access to capable large language models. They must understand company-specific definitions of revenue, customer churn, or account health, know which information they can access, determine appropriate actions, and operate within monitoring and governance frameworks1

1

. "You may have the best of the model, but you won't be able to reliably reason and act on behalf of your enterprise" without that context, Kumar stated2

2

.

Research Shows Harness Impact on Performance

Salesforce's own AI researchers demonstrated that the technical infrastructure surrounding a model substantially affects agent performance. In a September 8 preprint research paper, Salesforce researchers define the harness as the system prompt, tool set, execution hooks and context-management scaffolding surrounding an LLM1

1

.

Across seven enterprise agent benchmarks, evolving the harness around a smaller Qwen model increased average task success rates from 29.2% to 78.0%—a 48.8 percentage-point improvement without changing underlying model weights2

2

. However, the research revealed complications. After optimizing the harness, researchers fine-tuned the weaker model to imitate trajectories from a stronger expert model. Average success fell from 78.0% to 63.1%, with performance declining across all seven tasks due to loss of model-harness fit1

1

.

Composable Architecture Built on Existing Investments

The Enterprise AI Harness builds on Salesforce's recent progress in Headless 360 to create a more composable architecture4

4

. Rather than requiring rip-and-replace, the system leverages existing customer investments in Data 360, Informatica, MuleSoft, Tableau, Agentforce, and the Salesforce Platform5

5

. This allows enterprises to transform existing data, integrations, workflows, business logic, metadata, analytics, permissions and governance into composable building blocks for AI agents4

4

.

The AI Control Plane will enable administrators to centrally monitor AI agent performance and inference costs while ensuring adherence to cybersecurity policies3

3

. Kumar describes it as an evolution of API management tooling Salesforce already sells through MuleSoft, extended to manage MCP servers, LLM endpoints and an agent registry4

4

.

Intelligent Model Routing and Cost Management

Companies often use multiple LLMs in AI agent projects to optimize costs—frontier models power complex tasks while simpler agents use smaller, cheaper models3

3

. The Enterprise AI Harness will include intelligent model routing to automatically send agent requests to the most suitable LLM based on parameters such as accuracy and cost3

3

.

Kumar acknowledges gaps in the initial formulation, particularly around governance and cost controls. "I wouldn't say we have all the capabilities, just to be fair, because some of these things, especially when it comes to trusted security and the FinOps piece, there are things that we need to go build," Kumar stated4

4

. Salesforce is actively working on these aspects through research and development with customers to better align costs and enterprise value4

4

.

Watch how enterprises balance the need for AI-driven enterprise operations with practical cost management as Salesforce builds out these capabilities through early 2027.