ServiceNow CoreAI Launches AutoSynthData for Training Enterprise Agents

On October 2, 2026, ServiceNow CoreAI introduced AutoSynthData, a tool designed to generate synthetic training data for enterprise agents. In testing within the EnterpriseOps Gym, AutoSynthData produced 2,000 synthetic samples for the Hybrid domain in just 18 hours, achieving a mean Pass@1 improvement of 7.2 percentage points, which corresponds to a relative improvement of 35%. The verifier success rate in this domain increased from 63.01% to 68.55%, effectively closing 59% of the original Pass@1 gap.

In the IT Service Management (ITSM) domain, AutoSynthData generated 1,994 synthetic samples over a longer duration of 66 hours, resulting in a mean Pass@1 increase from 18.77% to 27.18%. This demonstrates the tool's capability to enhance model performance across different operational environments.

ServiceNow CoreAI emphasized that the effectiveness of training tasks is contingent upon the specific environment and the model being utilized. This development follows previous advancements in synthetic data generation and AI training methodologies, indicating a continued trend towards optimizing enterprise AI solutions.