What happened
Salesforce announced specific rollout dates for new capabilities in its Agent Fabric platform, including Agent Evals, API Evals, and a Triggered-Agent Kill Switch. The company set a November beta and January 2027 general availability target for these tools, which are designed to monitor, test, and automatically suspend malfunctioning AI agents across various enterprise environments.
Salesforce detailed the next phase of its Agent Fabric platform on October 27, focusing on evaluations that identify malfunctioning AI agents and initiate corrective actions. The announcement sets a November beta and January 2027 general-availability target for Agent Evals, a tool first previewed at the Dreamforce conference in September. This tool is designed to generate tests, measure outcomes against success criteria, and connect failures to remediation processes.
A key addition is the Triggered-Agent Kill Switch, scheduled for a January 2027 release. While a basic kill switch became generally available in September, this new feature connects detected problems to policy-driven intervention. It allows organizations to automatically quarantine agents that fall outside defined behavioral or performance thresholds, such as an agent that begins producing incorrect answers or consuming excessive resources.
The announcement also includes the launch of a Custom Agent Scanner beta, intended to extend agent discovery to internal systems and environments behind firewalls. This complements existing discovery tools and allows administrators to build a universal registry of agents across different platforms without requiring customers to relocate their agents to Salesforce's own Agentforce platform.
Salesforce also highlighted expanded cost controls, including a FinOps Dashboard and Model Wallet budgets. These tools allow organizations to set dollar or token limits by provider or model. The company claims these optimization techniques, such as semantic caching and payload compression, can reduce token usage by up to 75% in test scenarios, though these are vendor-reported figures rather than guaranteed customer results.
Source details: venturebeat.com
Why it matters
This update addresses a critical gap in enterprise by linking performance monitoring directly to automated corrective actions. By providing a unified registry and evaluation framework that works across competing platforms, Salesforce aims to reduce the operational risk and cost unpredictability associated with deploying multiple AI agents. This is significant for organizations that need to enforce strict safety and budget policies without migrating their existing agent infrastructure to a single vendor's ecosystem.
The central value of this update is the reduction of the distance between discovering an AI agent problem and addressing it. Previously, enterprises often had to manually identify the owner of a malfunctioning agent, navigate different platforms, and obtain permissions to suspend it. Agent Fabric aims to automate this workflow, providing a more responsive governance layer for complex AI environments.
By emphasizing that Agent Fabric is not Salesforce-specific, the company is positioning itself as a neutral governance layer for the multi-vendor AI market. This is a significant strategic move, as it allows enterprises to maintain their existing agent investments while adding a layer of oversight and control that spans across different frameworks and providers.
The integration of evaluation and enforcement is particularly important for . As agents become more autonomous, the ability to automatically trigger a shutdown based on real-time performance metrics is a critical safety feature. This moves beyond static security controls to dynamic, behavior-based governance.
Interactive Mechanism: How It Actually Works
Explore the underlying technology behind this development interactively.
crm_get_transaction(id='4092').An agent must create a draft calendar event for Tuesday at 2 p.m. Which evidence would establish the requested result?
What to watch next
The primary focus should be on the November beta launch to assess the practical effectiveness of the evaluation triggers. Additionally, enterprises should monitor how the Triggered-Agent Kill Switch performs in production environments, specifically regarding its ability to scope interventions to specific requests or tenants without disrupting broader system operations.
The November beta of Agent Evals and API Evals will be the first real-world test of these capabilities. Enterprises should watch for how well the evaluation tools handle edge cases and whether the automated remediation steps are reliable enough to be trusted in production environments.
The January 2027 release of the Triggered-Agent Kill Switch will determine if Salesforce can successfully connect its evaluation results to enforceable gateway controls. The effectiveness of this feature will depend on the breadth of supported integrations and the granularity of the intervention options available to administrators.
Pricing and access conditions for the specific Agent Fabric configuration remain somewhat opaque. While the broader MuleSoft portfolio starts at $2,000 per month, the cost for specific evaluation workloads is not publicly listed. Buyers should monitor for more detailed pricing information as the beta phase progresses.