Salesforce Inc. today previewed two offerings that will help customers build and manage artificial intelligence agents.
Developers turn a large language model into an agent by extending it with various add-ons. Those add-ons can include prompts, database connectors and a range of other technical assets. The technical assets used to customize an agent are collectively known as a harness.
The first offering that Salesforce debuted today is called the Enterprise AI Harness. It’s a collection of technologies designed to improve the reliability and security of customers’ AI agents. According to the company, many of the technologies that underpin the offering are already available in its cloud services. The rest will start rolling out early in Salesforce’s 2028 fiscal year, which begins next February.
The cloud giant says that Enterprise AI Harness will comprise six main feature collections. Half of them focus on the business data that AI agents use to carry out their work.
The first feature collection, Trusted Context, brings together data assets such as customer records and product availability information. Salesforce says there will also be support for “real-time signals.” In sales, a real-time signal is usually a buyer action that requires a company to quickly adjust its outreach efforts.
Another set of Enterprise AI Harness features will ensure that the data used by AI agents is accurate. Salesforce plans to provide multiple information quality controls including a data lineage tool. Cybersecurity teams, in turn, will gain the ability to regulate which agent can use what dataset and how.
Companies often use multiple LLMs in AI agent projects to optimize costs. A frontier model powers agents that work on complex tasks while simpler agents send requests to a smaller, cheaper model. The Enterprise AI Harness will include an “intelligent model routing” mechanism designed to ease that use case. It will automatically send agent requests to the most suitable LLM based on parameters such as accuracy and cost.
LLMs are nondeterministic, which means that entering the same prompt five times can result in five different answers. According to Salesforce, Enterprise AI Harness will enable customers to pair their AI agents with deterministic reliability controls. It will also ease the task of connecting agents to external applications.
The other offering that Salesforce previewed today is called the AI Control Plane. The company says that it will enable administrators to centrally monitor AI agents’ performance and inference costs. Additionally, the offering will help ensure that agents adhere to cybersecurity policies.
“Models will continue to change, and intelligence will increasingly be available everywhere,” said Rohan Kumar, Salesforce’s president and chief platform and engineering officer. “What will differentiate an enterprise is the trusted, proprietary context it brings to that intelligence — starting with the customer — and its ability to securely turn that context into action. That’s what we’re building with Salesforce’s Enterprise AI Harness.”
