Salesforce unveils its Trusted Enterprise AI Harness to enhance agent governance and streamline enterprise AI initiatives.
As enterprises increasingly experiment with artificial intelligence, many find themselves tangled in a web of pilot projects, unable to emerge into full operational capacity. Recent data reveals that a staggering 88% of organizations looking to integrate agent initiatives into their operations never see these projects reach production. Additionally, research from Gartner indicates that over 40% of agentic AI projects fall victim to cancellation due to rising costs and insufficient acquisition-in-sweden/">institutions/">risk management.
In response to this challenge, Salesforce has announced a strategic initiative designed to address these pressing issues within the enterprise AI landscape. On September 10, 2026, the company unveiled its Trusted Enterprise AI Harness, a bold move that aims to establish Salesforce as the leading management layer for agentic AI across all enterprises, including those that utilize AI agents outside of the Salesforce ecosystem.
The Trusted Enterprise AI Harness is structured around six fundamental capabilities: Trusted Context, Trusted Agency, Trusted Action, Trusted Governance, Trusted Security, and Trusted Models. At the heart of this initiative is a novel AI Control Plane that enables businesses to discover and register agents, enforce identity and policy, oversee their life cycles, evaluate their performance, and ultimately control costs.
Significantly, the AI Control Plane supports the management of both Salesforce-native and third-party AI agents. This flexibility is instrumental for organizations aiming to create cohesive agent workflows across diverse platforms, as highlighted by Rohan Kumar, Salesforce's President and Chief Platform and Engineering Officer. He believes that the competitive advantage lies not in the AI model itself, but in the tailored context that organizations can inject into these models.
"The Agentic Enterprise won’t be defined by which model a company chooses. Models will continue to evolve, and intelligence will become ubiquitous. What differentiates an enterprise is the secure, proprietary context it applies to that intelligence – fundamentally rooted in customer needs – and its capacity to translate that context into actionable insights," Kumar remarked.
This significant announcement follows shortly after Salesforce’s partnership with Anthropic through the Claudeforce initiative, where Claude was designated as the default reasoning engine for Salesforce's platform. By reinforcing both the reasoning engine layer and the new governance framework, Salesforce seeks to dominate the agentic operations space comprehensively.
A growing demand for effective management in this area is evident. A VentureBeat Intelligence survey conducted in July 2026 indicates that 85% of enterprises are managing two or more agent orchestration platforms, with an average of 3.1 platforms per organization. This fragmentation is precisely the issue that Salesforce aims to alleviate by offering a centralized control mechanism. The company's strategy is based on the premise that businesses will be motivated to pay for the consolidation of their agent management processes.
Nonetheless, this ambitious proposition creates a complex dynamic for buyers. Although Salesforce promotes an open architecture – facilitating multi-channel communication, application programming interfaces (APIs), Skills, and plug-ins – the underlying objective is unmistakably geared towards taking charge of the management layer across various AI agents. Shawn Malhotra, CTO of Rocket Mortgage, emphasizes the need for flexibility in this swiftly evolving landscape, stating, "AI is progressing at a speed that surpasses any technology we’ve encountered, which is why composability is crucial for us. We don’t want to tie our future to a single, closed stack; we require the liberty to adapt as the market transforms. That’s what appealed to us about Salesforce’s new Enterprise AI Harness."
The core challenge for decision-makers lies in balancing the allure of composability against the potential risks of vendor lock-in. By centralizing governance within one platform, organizations must trust that this framework will remain neutral and capable of interoperability as the industry's landscape changes.
Despite the promising features of the Trusted Enterprise AI Harness, it is important to note that general availability is not expected until early fiscal year 2028, beginning in February 2027. As of now, Salesforce has not revealed any pricing details for this service. This announcement appears to be strategically timed to create momentum as the company gears up for Dreamforce, outlining its vision for the future of agentic enterprise operations.
For enterprises considering this platform, the focus will likely shift toward how effectively Salesforce can integrate third-party agents into its Control Plane. The success of this endeavor will hinge on whether Salesforce can genuinely bridge the existing chasm between pilot projects and live production systems, or if it will instead introduce another layer of complexity to an already fragmented infrastructure.
The landscape of enterprise AI agents, workplace automation, and agentic tools is evolving at an unrelenting pace. Organizations must remain agile and informed as they navigate this transformation in automation and productivity.