Salesforce Agent Fabric Evolves for AI Oversight

Salesforce is detailing the next phase of its Agent Fabric platform, introducing new evaluations designed to identify malfunctioning AI agents and initiate corrective actions. Evaluations for agents are scheduled for a November beta, with general availability targeted for January 2027. This platform enhancement also includes expanded discovery capabilities and existing spending and governance controls.

The forthcoming API Evals tool will test the software connections that agents depend on, with both evaluation products set for beta in November and general availability in January 2027. A key addition is the Triggered-Agent Kill Switch, slated for January 2027, which connects detected problems to policy-driven intervention.

Existing agent discovery tools are being extended with a Custom Agent Scanner, entering beta now and planned for general availability in January 2027. This scanner aids in discovering agents across internal systems and environments behind firewalls. Spending controls are also highlighted, including a FinOps Dashboard and Model Wallet budgets, building on features outlined over the summer.

Salesforce is also emphasizing Open Policies for applying governance across federated gateways. The core message is to shorten the distance between discovering an AI agent problem and addressing it, enabling enterprises to manage agents effectively, regardless of where they are built.