Work Agents: Systems, Capabilities, and Optimization

Zishan Xu^1*, Enjun Du^2,3*, Mingqi Gao^4*, Cheng Yang³, Junhao Shen⁵, Zirong Chen⁴, Zhiyuan Yao⁶, Yifu Guo⁷, Yuquan Lu⁷, Jinyang Huang⁸, Zhengxi Lu⁶, Siyi Liu³, Ziyu Zheng², Yongqi Zhang^3†, Difan Zou^2†, Weiwen Liu^1†

¹Shanghai Jiao Tong University   ²The University of Hong Kong   ³The Hong Kong University of Science and Technology (Guangzhou)   ⁴Tsinghua University
⁵The Chinese University of Hong Kong   ⁶Zhejiang University   ⁷Sun Yat-sen University   ⁸Central South University

^*Equal contribution   ^†Corresponding authors

Work Agents are AI systems that combine foundation models with runtime harnesses to carry out delegated digital work in stateful workspaces. They use tools, interact with applications, and coordinate workflows to produce answers, editable artifacts, and changes to workspace state.

This repository is the companion reading list for Work Agents: Systems, Capabilities, and Optimization. It brings together research across office productivity, enterprise operations, software engineering, scientific research, and computer use:

• Agents, methods, and systems — architectures, training methods, environments, memory, and agent harnesses.
• Benchmarks and evaluation — task suites and evaluation methods for digital work.
• Surveys and resources — foundational studies, open-source tools, and official documentation.

Read the survey on arXiv, and visit the project homepage for the project overview.

• Agents, Methods & Systems
• Benchmarks & Evaluation
• Surveys & Foundations
• Resources
• Citation
• Contributing

Agents, learning methods, training environments, and runtime systems for producing work outcomes and improving capability. 278 entries.

Benchmarks, datasets, evaluation environments, and protocols for measuring task completion, work quality, reliability, and cost. 190 entries.

Research syntheses, foundation-model reports, conceptual foundations, reasoning and verification studies, and empirical work on productivity and collaboration. 92 entries.

Official product pages, engineering reports, documentation, evaluation infrastructure, and project resources. These entries are excluded from the research-paper count. 59 entries.

If you find this survey or reading list useful, please cite:

@misc{workagents2026,
  author       = {Zishan Xu and Enjun Du and Mingqi Gao and
                  Cheng Yang and Junhao Shen and Zirong Chen and
                  Zhiyuan Yao and Yifu Guo and Yuquan Lu and
                  Jinyang Huang and Zhengxi Lu and Siyi Liu and
                  Ziyu Zheng and Yongqi Zhang and Difan Zou and
                  Weiwen Liu},
  title        = {Work Agents: Systems, Capabilities,
                  and Optimization},
  year         = {2026},
  eprint       = {2610.35845},
  archivePrefix = {arXiv},
  primaryClass = {cs.LG},
  url          = {https://arxiv.org/abs/2610.35845}
}

Contributions are welcome! Open an issue or submit a pull request to suggest papers, resources, or corrections.