Salesforce has announced the addition of Agent Evals, API Evals, and enhanced cost control features to Agent Fabric, its AI agent control plane designed to oversee and govern AI agents across various systems.
The company said these updates are intended to help enterprises intervene before AI agent failures or escalating costs affect customers.
Go deeper with GlobalData
Access deeper industry intelligence
Experience unmatched clarity with a single platform that combines unique data, AI, and human expertise.
Agent Fabric currently processes more than 1.2 billion requests each day, according to Salesforce.
As organisations integrate more AI agents, the risk of unnoticed failures or increasing expenses also grows, particularly given that many issues go undetected in production environments.
The company noted that fewer than half of organisations with AI agents evaluate their agents’ performance on real-world traffic, leaving potential problems unaddressed until they result in customer complaints or budget overruns.
The new Agent Evals capability automatically creates and runs tests before and during agent deployment, assessing agents, large language models (LLMs), and Model Context Protocol (MCP) servers against set quality benchmarks.
When these benchmarks are not met, the system can trigger interventions, such as activating an Agent Kill Switch, according to company policy.
API Evals provides automated testing of APIs relied upon by agents, aiming to identify any breaking changes before they enter production.
Salesforce has also introduced advanced cost controls through features like the FinOps Dashboard and Model Wallet.
These tools are designed to provide a consolidated view of AI spending, enforce strict usage limits, and reduce token consumption by up to 75% in testing with techniques such as semantic caching and response trimming.
Salesforce MuleSoft and Agent Fabric senior vice president and general manager Andrew Comstock said: “Enterprises can only scale agents as fast as they can keep them accurate and on budget. Today, spotting a wrong answer or a spending spike is one job, and stopping it is another.
“Agent Fabric unifies them, bringing evaluations, cost controls, and intervention together so teams can contain issues instantly and scale agents with confidence.”
Additional enhancements include the Custom Agent Scanner, which allows for the discovery and cataloguing of agents, models, and APIs across internal or third-party environments.
Open Policies for Federated API Gateways enable policy enforcement across multiple API management systems.
Salesforce states these updates support continuous agent evaluation and governance, helping organisations maintain control as they scale up the deployment of AI agents.
Earlier this month, Salesforce agreed to acquire US cloud software company Braze in a deal valued at approximately $5.66bn.
