Salesforce Adds Kill Switch and Spending Caps to Agent Fabric Platform

Quick Facts

• Salesforce made its FinOps Dashboard, Model Wallet, and Agent Kill Switch generally available on Oct. 27, 2026, with automated shutdown triggers and evaluation tools arriving in January 2027.
• Agent Fabric handles 1.2 billion daily requests and covers agents built on AWS Bedrock, Azure AI Foundry, Google Vertex AI, Microsoft Copilot, and custom frameworks.
• Cost Optimization Policies cut token consumption by up to 75% in test scenarios through semantic caching, payload compression, and response trimming.

Salesforce expanded Agent Fabric on Oct. 27 with new evaluation tools and cost controls designed to catch failing AI agents before they damage production workflows, according to the company's announcement.

The release makes the FinOps Dashboard, Model Wallet, Cost Optimization Policies, and a basic Agent Kill Switch generally available today. Agent Evals and API Evals enter beta in November 2026. Both evaluation products and a triggered, automated version of the kill switch reach general availability in January 2027.

What Agent Fabric Does

Agent Fabric is a vendor-agnostic control plane. It finds every agent, Model Context Protocol server, and API a business runs, then governs permissions, connects systems, and tracks costs. It does not require customers to move agents to Salesforce infrastructure.

Andrew Comstock, SVP and GM of MuleSoft and Agent Fabric, described the problem the product addresses: "I have too many of them. The technologies are totally different. Nothing talks to each other. Everything runs in a silo." Salesforce data shows 96% of enterprise organizations now run autonomous agents across multiple platforms, while 58% struggle to maintain consistent security and trust standards across those platforms.

New Evaluation Tools

Agent Evals automatically generate tests before deployment, then run agents, large language models, and MCP servers against defined quality thresholds in production. When results fall short, the system triggers predefined interventions. MuleSoft first previewed Agent Evals at Dreamforce in September 2026.

API Evals complement this by generating performance tests directly from API specifications. The tool flags breaking dependency updates before they reach production workflows.

Poonam Deshpande, Director of Engineering at ASICS Digital, said Agent Evals go beyond detection: "When an agent's quality drops, Agent Fabric can step in automatically, alerting the right owner or pulling the agent offline before a small issue becomes a widespread customer problem."

The Kill Switch

The basic Agent Kill Switch is available now and lets teams take a failing agent offline immediately. The triggered version, which fires automatically when an evaluation fails, ships in January 2027. Salesforce advises basing purchasing decisions on currently released capabilities, not the January roadmap.

MuleSoft's platform currently processes 1.2 billion daily requests through Agent Fabric. Fewer than half of businesses with agents evaluate them on live traffic, according to Salesforce's own figures, which the company cited as the core gap Agent Evals addresses.

Cost Controls

The Model Wallet enforces spending limits per agent, per instance, or per business unit. It tracks spend in real time and blocks new requests before costs exceed budget rather than after the invoice arrives. Forecasting features flag projected overruns days in advance.

The FinOps Dashboard gives budget owners a single view of AI spending across the organization. Cost Optimization Policies reduce token usage through semantic caching, payload compression, and response trimming.

Adam Jones, Executive Director of Enterprise Transformation at GE Appliances, said the visibility gap was the primary concern: "A lot of what's going on in big organizations with AI today involves one-offs, lacking governance or visibility. With Agent Fabric, we now have that single control plane where we can see our agents, who built them, and shut them down if something goes wrong."

Comstock framed the release around a scaling constraint that applies across the industry: "Enterprises can only scale agents as fast as they can keep them accurate and on budget."