Salesforce has added a kill switch, evaluation tools and hard spending limits to Agent Fabric, the MuleSoft control plane it sells for managing AI agents across platforms.
Agent Fabric now handles more than 1.2 billion requests a day, and Salesforce says the gap it is closing is that agents fail silently, giving wrong answers and running up token costs before anyone notices. Its research puts 96% of enterprises running agents on more than one platform, and it cites a LangChain survey in which fewer than half of organisations with agents evaluate them on live traffic.
Agent Evals generate tests before deployment, then assess agents, models and MCP servers in production against quality thresholds. When results slip, predefined interventions fire, including the Agent Kill Switch, with human approval where policy requires it. API Evals build functional and performance tests from API specifications to catch breaking changes. On cost, a FinOps Dashboard and Model Wallet give budget owners one view of AI spend and limits that stop requests before a budget is exceeded, while Cost Optimization Policies use semantic caching, payload compression and response trimming to cut token use by up to 75% in Salesforce's test scenarios.
Two further updates extend discovery. A Custom Agent Scanner catalogues agents, models, MCP servers and APIs in internal environments and third-party platforms, and Open Policies for Federated API Gateways applies one policy, such as rate limiting, across gateways from Azure, Apigee and Kong in a single action.
Andrew Comstock, SVP and GM of MuleSoft and Agent Fabric, said: "Agent Fabric unifies them, bringing evaluations, cost controls, and intervention together so teams can contain issues instantly and scale agents with confidence." Poonam Deshpande, Director of Engineering for Digital Commerce at ASICS Digital, a beta participant, said of Agent Evals: "That’s the missing piece for running agents at scale, and it will give us the confidence to keep expanding AI across our business."
On availability, the Agent Kill Switch, Open Policies, FinOps Dashboard, Model Wallet and Cost Optimization Policies are generally available now. Agent Evals and API Evals enter beta in November 2026 with general availability planned for January 2027, the same month the triggered version of the kill switch is due. The Custom Agent Scanner is in beta today.
To stay across the latest in cloud, AI and enterprise tech analysis from Compare the Cloud, subscribe to our weekly newsletter at https://www.comparethecloud.net/newsletter
