Salesforce details Agent Fabric next phase: agent evals, API tests and triggered kill switch

Salesforce on October 27 detailed the next phase of its Agent Fabric platform, adding evaluation capabilities designed to detect malfunctioning AI agents and trigger corrective action, along with broader discovery options and previously announced governance and spending controls.

Agent Evals, first unveiled by MuleSoft at Dreamforce in September, is scheduled to enter beta in November and reach general availability in January 2027. The companion API Evals tool, which generates functional and performance tests from API specifications, is also expected in beta in November and GA in January. Evaluations can extend beyond pre-deployment testing to check agents, language models and Model Context Protocol servers in production, identifying behavior outside an organization's thresholds.

A Custom Agent Scanner entered beta on October 27, extending discovery to internal systems, firewall-protected environments and additional platforms, with GA planned for January. Salesforce also placed a triggered-agent kill switch on the January roadmap, linking detected issues to policy-driven intervention. The basic kill switch became generally available in September, using gateway controls to block governed traffic with scoped interventions.