San Francisco-based Salesforce announced additions to Agent Fabric on Tuesday, Oct. 27. Agent Fabric is the MuleSoft-built control plane companies use to find, govern and monitor AI agents across platforms. Salesforce says it handles more than 1.2 billion requests a day.
What's new
- Agent Evals: Generates tests before deployment, then checks agents, LLMs and MCP servers in production against quality thresholds. When an agent falls short, it can trigger predefined responses, including the Agent Kill Switch, with human approval if a company's policy requires it.
- API Evals: Builds tests from API specifications to catch breaking changes before agents that depend on those APIs fail.
- Cost controls: The FinOps Dashboard and Model Wallet set hard spending limits. Salesforce says Cost Optimization Policies can cut token use by up to 75% in its test scenarios.
- Custom Agent Scanner and federated gateway policies extend discovery and policy enforcement to internal systems and gateways from Azure, Apigee and Kong.
When
Agent Evals and API Evals enter beta in November, with general availability in January 2027. The manual Agent Kill Switch, cost tools and federated gateway policies are available now. The evaluation-triggered kill switch reaches general availability in January 2027.
ASICS, Capita, GE Appliances and Newell Brands provided customer quotes. Unite.AI points to outside survey data showing that fewer than half of organizations with agents evaluate them on live traffic.