The work is illustrated with EnterpriseOps Gym, using a released dataset, in a Hugging Face article published October 2, 2026.
AutoSynthData identifies capability gaps from evaluation runs of a target model. A stronger teacher is used to characterize solvable failures, after which the system creates new tasks with a system specification, user prompt, and verifier. The generated tasks vary prompts, environment states, entities, workflows, and tool combinations rather than retaining the original evaluation prompts or trajectories.
Candidates are executed and checked before acceptance. The stated configuration favors tasks solved by the target model in no more than one of three trials and by a stronger solver in at least two of three. It also applies positive and negative verification, bounded repair attempts, and batch-level review for coverage and diversity.
The article describes a method and an experiment, not a generally available ServiceNow product. It does not establish broad performance gains, deployment results, pricing, or availability for customer environments.