Salesforce Expands Agent Fabric to Automatically Stop Runaway Costs and AI Agent Failures Before They Impact Customers

Salesforce unveiled Agent Evals, API Evals, and advanced cost controls for Agent Fabric, an AI control plane designed to manage and govern agents across multiple platforms. The new tools aim to address the challenge of silent agent failures that can lead to incorrect responses and escalating costs, which often go unnoticed until they impact customers. By providing continuous evaluation and automated interventions, Agent Fabric helps organizations catch issues early and prevent widespread disruptions.

Agent Fabric now offers automated discovery and governance of agents across major AI platforms, with extensions to internal, third-party, and API environments. The platform enables predefined responses to quality or cost deviations, reducing manual oversight and ensuring agents operate as intended after deployment. Enterprises can set hard spending limits and receive alerts when agent performance or costs deviate from expectations.

The announcement highlights the need for proactive oversight as fewer than half of organizations evaluate agents on live traffic. Agent Evals and cost controls are designed to bridge this gap by continuously assessing agent performance, tool usage, efficiency, and security compliance. This unified approach allows teams to scale AI agents confidently while maintaining accuracy and budget adherence.

Customer feedback underscores the practical value of Agent Fabric. Companies like ASICS Digital, Capita, GE Appliances, and Newell Brands emphasize the platform’s role in centralizing agent governance, improving visibility, and enabling responsible AI deployment. The tools are positioned as critical for transitioning from experimental AI projects to enterprise-scale, trustworthy automation.