Salesforce expands Agent Fabric to catch agent failures and cap AI costs
Agent Evals, API Evals and FinOps tools aim to stop broken agents and runaway spend before customers notice.
Salesforce is adding Agent Evals, API Evals and advanced cost controls to Agent Fabric, its AI control plane. The aim is to automatically stop AI agent failures and rein in AI usage costs before they reach customers.
The new features include continuous evaluation, automated intervention, a FinOps dashboard and “model wallets” for budget management. They roll into beta and general availability between November 2026 and January 2027.
Why it matters: unpredictable AI spend is a top worry for many businesses. Measuring each agent and giving it a budget is a sensible pattern for anyone putting agents into sales or service workflows.