WorkFactory: Turning Professional Intents into Executable Work for Scalable Agent Training
Summary
WorkFactory introduces a scalable framework for constructing coherent, executable professional tasks via information-level planning and repair, producing 112,686 tasks across 612 occupations to fine-tune the WorkPilot model family (Qwen 4B/27B/35B-A3B), which improve over base models on professional-work benchmarks.
View Cached Full Text
Cached at: 10/30/26, 03:15 AM
Paper page - WorkFactory: Turning Professional Intents into Executable Work for Scalable Agent Training
Source: https://huggingface.co/papers/2610.37161 Authors:
,
,
,
,
,
,
,
,
,
,
,
Abstract
WorkFactory builds coherent professional tasks via information planning and repair, yielding WorkPilot models that improve expert-level agent performance across benchmarks.
Scalingsupervised trainingfor working agents requires diverse professional tasks that yield reliable expert demonstrations. However, missing or inconsistent task information can obstruct teacher execution and weaken the resulting supervision. We proposeWorkFactory, a scalable framework for constructing coherent and executable professional tasks. Starting from professional intents, information-level planning builds aninformation support graphto guide task instantiation, explicitly representing the dependencies and locations of information required for the intended outputs.Probe executionsthen identify support failures and guide repairs while preserving the intended work. We construct 112,686 tasks spanning 612 occupations and use the collected teacher trajectories tofine-tuneQwenmodels at 4B, 27B, and 35B-A3B scales, yielding theWorkPilotfamily. All three models improve over their respective base models across four professional-work benchmarks, withWorkPilot-35B-A3B improving from 879 to 1565Eloon GDPval-AA v2.1.Scaling analysesreveal model-dependent diminishing returns, with execution length first increasing and then decreasing. Controlled component analyses show that planning and repair reduce probe-diagnosed task environment defects and improve downstream performance. The data and models are available at https://github.com/IQuestLab/WorkFactory.
View arXiv pageView PDFGitHub5Add to collection
Get this paper in your agent:
hf papers read 2610\.37161
Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash
Models citing this paper0
No model linking this paper
Cite arxiv.org/abs/2610.37161 in a model README.md to link it from this page.
Datasets citing this paper0
No dataset linking this paper
Cite arxiv.org/abs/2610.37161 in a dataset README.md to link it from this page.
Spaces citing this paper0
No Space linking this paper
Cite arxiv.org/abs/2610.37161 in a Space README.md to link it from this page.
Collections including this paper0
No Collection including this paper
Add this paper to acollectionto link it from this page.
Similar Articles
Workflow-GYM: Towards Long-Horizon Evaluation of Computer-use Agentic tasks in Real-World Professional Fields
Workflow-GYM is a benchmark for evaluating AI agents on long-horizon GUI tasks in professional domains. Experiments show that even top models achieve only ~30% success, revealing significant challenges.
WorkRover: Scaling Long-Horizon Cowork Agent Training Via Recursive Environment Evolution
WorkRover is a framework for training long-horizon cowork agents through recursive environment evolution, introducing Factorial Compatibility Replay and Verifier-to-Turn Credit Assignment to improve training signals. Post-training Qwen3.5-35B-A3B into WorkRover-35B-A3B yields large gains on JobBench, OfficeQAPro, and GDPvalv2 benchmarks.
Scaling Verifiable Environments for Long-horizon Work Agents
论文提出 WorkForge，一个从真实世界资源构建可验证工作智能体训练环境的可扩展合成框架，构建了覆盖 40 个专业领域的 16.7K 个可验证环境，使 Qwen3.5-35B-A3B-Base 在 GDPVal 上从 45.5 提升至 73.6。该论文已被撤稿（withdrawn）。
OfficeTown: Generative Organizations for Evolving Agent Intelligence
OfficeTown 是一个生成式组织模拟框架，可通过模拟专业组织的工作流为 AI 代理生成工作导向的训练数据与评测基准。作者构建了 OfficeTown Bench（406 个任务、4836 条评测标准），并表明用该数据进行监督微调和 on-policy 蒸馏能显著提升 Qwen3.5-9B 在 GDPval 等外部专业工作基准上的表现。
WorkBench Revisited: Workplace Agents Two Years On
This paper revisits the WorkBench benchmark for workplace agents two years after its initial release, showing that the best agent (Claude Opus 4.8) now completes 89% of tasks with only 2.5% harmful side effects, compared to GPT-4's 43% completion and 26% harm rate in 2024. It finds that capability and safety improve together, open-weight models have drastically lowered costs, and some basic mistakes persist.