Salesforce Aims to Centralize Agentic AI Management with New Harness Platform
Salesforce has announced its Trusted Enterprise AI Harness, a platform designed to manage agentic AI across entire enterprises. The company is attempting to solve the problem of pilot projects that never reach production by formalizing the agent governance stack.
Data from IDC and Lenovo suggests that 88% of enterprises with agent initiatives never ship to production, while Gartner reports that over 40% of agentic AI projects are cancelled due to escalating costs and inadequate risk controls. Salesforce is betting that its platform will reduce this production gap by offering a centralized control plane.
The Harness consists of six core capabilities: Trusted Context, Trusted Agency, Trusted Action, Trusted Governance, Trusted Security, and Trusted Models. Central to this is a new AI Control Plane, which allows organizations to discover and register agents, establish identity and policy, manage lifecycles, evaluate performance, and control costs.
The success of this initiative depends on whether Salesforce can integrate third-party agents into the Control Plane without adding complexity to an already fragmented stack. Rohan Kumar, Salesforce's President and Chief Platform and Engineering Officer, argues that companies will be differentiated by their proprietary context, not the model they choose.