Salesforce announces new tools to audit AI agent quality and cost
Salesforce promises to catch agent problems before they turn into customer complaints; the news is based on the company's own announcement.
What's happening
According to Salesforce News, Salesforce has announced Agent Evals, API Evals, and advanced cost controls for Agent Fabric, its AI control plane for managing and governing agents across platforms. The company says automated interventions kick in when quality slips, and hard spending limits keep usage within budget. Agent Evals will enter beta in November 2026 and become generally available in January 2027.
Whose desk it lands on
The executive responsible for AI agents. When an agent's quality drops or its spending exceeds the budget, who signs off on stopping it, and at what threshold?
Our view
The CX Brief's interpretation, not a claim made by the source.
The announcement's core premise is that agent degradation should be caught by testing, not by customer complaints. In Salesforce's own scenario, a retailer switches to a cheaper model to cut costs, its agent starts giving inaccurate information about delayed orders, and scheduled tests catch the problem. That is why cost-saving decisions and customer experience decisions have to be made together.
