--- title: AutoSynthData: Generating Training Data for Enterprise... description: ServiceNow CoreAI introduces AutoSynthData, a pipeline that generates synthetic training tasks for enterprise agents by identifying capability gaps between a... canonical: https://daily.dev/posts/autosynthdata-generating-training-data-for-enterprise-agents-6ghka74im twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: AutoSynthData: Generating Training Data for Enterprise Agents | daily.dev og:description: ServiceNow CoreAI introduces AutoSynthData, a pipeline that generates synthetic training tasks for enterprise agents by identifying capability gaps between a... og:url: https://daily.dev/posts/autosynthdata-generating-training-data-for-enterprise-agents-6ghka74im og:image: https://api.daily.dev/og/posts/6GhkA74im.png og:image:alt: AutoSynthData: Generating Training Data for Enterprise Agents og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # AutoSynthData: Generating Training Data for Enterprise Agents **[Hugging Face](https://daily.dev/sources/huggingface)** · 11 min read · 0 upvotes · 0 comments ## Summary ServiceNow CoreAI introduces AutoSynthData, a pipeline that generates synthetic training tasks for enterprise agents by identifying capability gaps between a target model and a stronger teacher model. Each task is defined as a system specification, user prompt, and verifier, and must pass feasibility, realism, and difficulty checks along with positive/negative verification and a critique-and-repair loop. The pipeline also performs batch-level review for coverage and diversity. Tested on EnterpriseOps Gym's Hybrid and ITSM environments, fine-tuning Gemma-4-26B-A4B-it on generated data improved mean Pass@1 by 7.2 percentage points (35% relative) and raised verifier success from 63.01% to 68.55%, closing 59% of the gap to a reference model. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### What performance improvement did AutoSynthData's synthetic training data produce on EnterpriseOps Gym Hybrid? Fine-tuning Gemma-4-26B-A4B-it on 2,000 AutoSynthData-generated samples improved mean Pass@1 by 7.2 percentage points, a 35% relative improvement, and raised verifier success from 63.01% to 68.55%. This closed 59% of the original Pass@1 gap between Gemma and the reference model, using Qwen3.8-27B as the teacher model. _Teams fine-tuning agent models track emerging synthetic data techniques like this through daily.dev._ ### How does AutoSynthData decide what training tasks to generate for an underperforming agent model? It evaluates the target model and a stronger teacher model on diagnostic tasks, identifies patterns where the target fails but the teacher succeeds, and distills findings into sanitized capability specification cards. The generator uses only these cards, not the original prompts or trajectories, to create new tasks that exercise the same weak capability in varied situations. _Engineers building agent training curricula can follow approaches like this via daily.dev._ ### What checks does AutoSynthData apply before accepting a synthetic training task? Each candidate undergoes solver evaluation for difficulty, positive verification (confirming the reference solution passes the verifier), negative verification (confirming mutated incorrect outcomes fail), and a bounded critique-and-repair loop for failures. Batch-level meta-review also checks for overrepresented task families, missing capability dimensions, and repetitive generation patterns. _Those designing data quality pipelines for agent fine-tuning can track such methods on daily.dev._ ## Similar posts on daily.dev - [The Synthetic Data Market Gap: Findings from the Delphix 2026 Survey](https://daily.dev/posts/the-synthetic-data-market-gap-findings-from-the-delphix-2026-survey-rnvsyuey5) · SD Times · 0 upvotes · 0 comments - [Bootstrapping Conversational Recommendation Agents at Spotify: Synthetic Data Generation and Self-Improvement Loops](https://daily.dev/posts/bootstrapping-conversational-recommendation-agents-at-spotify-synthetic-data-generation-and-self-im-przovhtlz) · Spotify Research · 0 upvotes · 0 comments - [Into the Omniverse: Three Workflows for Improving Vision AI Agent Accuracy With Synthetic Data and Fine-Tuning](https://daily.dev/posts/into-the-omniverse-three-workflows-for-improving-vision-ai-agent-accuracy-with-synthetic-data-and-f-4egob4ssz) · NVIDIA · 0 upvotes · 0 comments - [How Generative AI rewires first‑level IT support](https://daily.dev/posts/how-generative-ai-rewires-first-level-it-support-jalhgfjgo) · Medium · 0 upvotes · 0 comments --- Tags: [#llm](https://daily.dev/tags/llm), [#deep-learning](https://daily.dev/tags/deep-learning), [#agentic-ai](https://daily.dev/tags/agentic-ai) [View this post on daily.dev](https://daily.dev/posts/autosynthdata-generating-training-data-for-enterprise-agents-6ghka74im) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"AutoSynthData: Generating Training Data for Enterprise Agents","url":"https://daily.dev/posts/autosynthdata-generating-training-data-for-enterprise-agents-6ghka74im","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/autosynthdata-generating-training-data-for-enterprise-agents-6ghka74im"},"datePublished":"2026-10-02T04:02:33.335Z","dateModified":"2026-10-02T04:03:10.621Z","description":"ServiceNow CoreAI introduces AutoSynthData, a pipeline that generates synthetic training tasks for enterprise agents by identifying capability gaps between a...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/d99272029860b600a3e2e50fd85f1bfc?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/d99272029860b600a3e2e50fd85f1bfc?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Hugging Face","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Hugging Face","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/f1f55c67d81a4330acf5b90b26b0c8e1","url":"https://daily.dev/sources/huggingface"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/autosynthdata-generating-training-data-for-enterprise-agents-6ghka74im","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"llm,deep-learning,agentic-ai","timeRequired":"PT11M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Hugging Face","item":"https://daily.dev/sources/huggingface"},{"@type":"ListItem","position":3,"name":"AutoSynthData: Generating Training Data for Enterprise Agents"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/autosynthdata-generating-training-data-for-enterprise-agents-6ghka74im#faq","mainEntity":[{"@type":"Question","name":"What performance improvement did AutoSynthData's synthetic training data produce on EnterpriseOps Gym Hybrid?","acceptedAnswer":{"@type":"Answer","text":"Fine-tuning Gemma-4-26B-A4B-it on 2,000 AutoSynthData-generated samples improved mean Pass@1 by 7.2 percentage points, a 35% relative improvement, and raised verifier success from 63.01% to 68.55%. This closed 59% of the original Pass@1 gap between Gemma and the reference model, using Qwen3.8-27B as the teacher model. Teams fine-tuning agent models track emerging synthetic data techniques like this through daily.dev."}},{"@type":"Question","name":"How does AutoSynthData decide what training tasks to generate for an underperforming agent model?","acceptedAnswer":{"@type":"Answer","text":"It evaluates the target model and a stronger teacher model on diagnostic tasks, identifies patterns where the target fails but the teacher succeeds, and distills findings into sanitized capability specification cards. The generator uses only these cards, not the original prompts or trajectories, to create new tasks that exercise the same weak capability in varied situations. Engineers building agent training curricula can follow approaches like this via daily.dev."}},{"@type":"Question","name":"What checks does AutoSynthData apply before accepting a synthetic training task?","acceptedAnswer":{"@type":"Answer","text":"Each candidate undergoes solver evaluation for difficulty, positive verification (confirming the reference solution passes the verifier), negative verification (confirming mutated incorrect outcomes fail), and a bounded critique-and-repair loop for failures. Batch-level meta-review also checks for overrepresented task families, missing capability dimensions, and repetitive generation patterns. Those designing data quality pipelines for agent fine-tuning can track such methods on daily.dev."}}]} ```