ServiceNow open-sources AutoSynthData for enterprise agent training

Summary

ServiceNow's CoreAI team published AutoSynthData, a pipeline that turns an agent's weaknesses into training data. It contrasts the target model's failures with a stronger teacher model's successes, generates and validates new tasks for those gaps, and shifts the curriculum toward what the model still finds difficult as it improves. The walkthrough uses the EnterpriseOps Gym environment and its released dataset.

Why it matters for our work

It shows, with a worked example, a path to training agents tailored to a company's own systems and rules rather than deploying off-the-shelf. Organizations working to raise agent performance on their own workflows can use it as a reference.

Translated from the Korean original. Summaries may be translated and edited. Commentary reflects our perspective; forecasts remain the source’s views.