Eigen RadarAI
Archive
Analysis

Salesforce schedules cross-platform agent test betas for November

Salesforce has set November beta and January 2027 general-availability targets for Agent Fabric’s agent and interface evaluation tools. A custom discovery scanner is entering beta, while automatically triggering an agent shutdown after a failed evaluation remains on the January roadmap. Existing spending and manual shutdown controls have a different availability status.

Artificial Intelligence··Morning
In an imagined workshop, a metal testing arm touches a cream-and-coral machine body and its clear connection; different sockets sit on a shelf behind.

Evaluation products receive beta and release targets

Salesforce has set a November beta and January 2027 general-availability target for Agent Evals and API Evals in Agent Fabric, its system for governing AI agents across platforms. The 27 October announcement gives the evaluation tools a rollout schedule. Shutdown initiated after evaluation also has a January 2027 general-availability target, distinct from the existing stop control.[1], [2]

Agent Evals is designed to generate tests before deployment and assess agents, language models and Model Context Protocol servers against defined quality thresholds during use. Those servers connect models to tools and information. Salesforce says results below a threshold can initiate predefined responses, including stopping an agent, with approval where organizational policy requires it.[1]

Interface tests target changes that break workflows

API Evals is intended to build functional and performance tests from application programming interface specifications. These interfaces let an agent communicate with other software. The proposed tests seek to detect changes that break a workflow before they reach production. Salesforce illustrates the process with a hypothetical retailer changing to a cheaper model, finding inaccurate order answers and retesting revised instructions. It presents a scenario rather than a measured customer result.[1]

Custom discovery enters beta ahead of evaluation release

The Custom Agent Scanner enters beta with the announcement and targets general availability in January 2027. It extends discovery to internal systems, environments behind firewalls and platforms outside standard integrations. Scanners can catalog agents, models, protocol servers and interfaces in a central registry. Existing spending controls and the manual Agent Kill Switch are listed as generally available. Automatically linking a failed evaluation to intervention remains part of the announced future rollout.[1]

References

  1. News sourceSalesforceSalesforce dates Agent Fabric evaluation betas for November↩1↩2↩3↩4
  2. News sourceAgentsCallingSalesforce separates available spending controls from future evaluation-triggered intervention↩
AI safety and regulationChip wars and the supply chain

More analysis from this desk

View all Brief analyses from the Artificial Intelligence desk →