[Digital Today reporter Chi-gyu Hwang] Salesforce has unveiled Enterprise AI Harness and an AI Control Plane to support building and managing AI agents, SiliconANGLE reported on Thursday (local time).

The report said developers add technologies such as prompts and database connectors to large language models (LLMs) to turn them into agents, and this bundle of add-on technologies is called a harness.

Salesforce's Enterprise AI Harness includes technologies designed to improve customer companies' AI agent reliability and security. Many of the technologies applied to the Enterprise AI Harness are already used in Salesforce cloud services, and the rest will be released in stages starting in the 2028 fiscal year beginning in February next year.

Enterprise AI Harness is made up of 6 feature groups. Half of them handle data used by AI agents for work. "Trusted Context" gathers data assets such as customer records and product inventory information. It also supports real-time signals. In sales, real-time signals refer to buyer behavior that companies need to respond to quickly.

There is also a feature group that ensures the accuracy of data used by AI agents. Salesforce plans to provide multiple quality management features, including a data lineage tracking tool. Security teams can control which agents can use which datasets and how.

Companies use multiple LLMs together in AI agent projects to reduce costs. A top-tier model handles complex tasks, while simpler tasks are passed to smaller, cheaper models. Enterprise AI Harness also includes an "intelligent model routing" feature that automatically assigns requests to the most suitable LLM based on criteria such as accuracy and cost.

Even if the same prompt is entered five times, an LLM can produce five different answers. Salesforce said Enterprise AI Harness offers deterministic reliability controls to compensate for such non-determinism. It also said it will make it easier to connect external apps and agents.

"AI Control Plane" helps administrators monitor the performance and inference costs of multiple AI agents in one place. It also includes functions to manage agents so that they follow security policies.

Rohan Kumm (로한 쿠마), Salesforce president and chief platform engineering officer, said, "Models will keep changing, and intelligence will become usable everywhere," adding, "For companies, differentiation comes from the ability to combine trusted, unique data, including customer information, with that intelligence and safely put it into action."