• [2026/07] 📃 We created this repository to share updates and related literature!

• 🧭 Overview
• 🪜 Stages of AI for Productivity
• 🌱 Nascent-diffusion Industries
  • 🏭 Manufacturing
  • 🌾 Agriculture
  • 🏠 Real Estate
  • 🏛️ Government and Public Administration
• 📈 Growing-diffusion Industries
  • 🛍️ Trade
  • 📰 Media
  • 🎓 Education
  • ⚖️ Law
• 🌍 Widespread-diffusion Industries
  • 💻 Information Technology
  • 💰 Finance
  • 🩺 Medical
• 📖 Citation
• 🤝 Contribution
  • 🌟 Star History
  • 👥 Contributors

We define AI for Productivity (AI4Productivity) as the use of Agentic AI in real workflows to deliver economically valuable work with less human effort. Our paper explains how agentic AI converts technical capability into sustained productivity through two mechanisms and four stages of delegated production responsibility:

• Two mechanisms of productivity gains — the Efficiency Mechanism, which reduces the time and labor required for existing work, and the Expansion Mechanism, which makes previously infeasible work economically viable.
• Four delegation stages — a progression from human-supervised conversational assistants to fully autonomous self-governing production systems.

Using this framework, we assess how agentic AI is currently embedded across nine industries, grouped into three levels of diffusion, and extract cross-sector insights on where productivity gains are visible, where they remain constrained, and what conditions enable deeper diffusion.

A four-stage framework for AI for Productivity evolution:

• Stage 1: AI as Conversational Assistant: Passive, single-turn informational tool, limited to preparatory assistance, fully dependent on human execution.
• Stage 2: AI as Reactive Operator: Automates narrow, repetitive tasks, substituting manual labor but requiring constant human oversight and error correction.
• Stage 3: AI as Adaptive Coordinator: Manages complex workflows, coordinates agents, and scales team-level capability, but relies on human-defined goals and requirements.
• Stage 4: AI as Self-Governing Production System: Autonomous ecosystem that translates abstract goals into actionable plans, orchestrates multi-agent systems, and expands the production frontier, with human oversight focused on strategic alignment and safety.

Nascent-diffusion industries represent the early stage of AI adoption, where Agentic AI primarily serves as a conversational assistant and reactive operator under human supervision.

In this phase, AI penetration remains limited, with applications focused on efficiency-driven, task-specific assistance rather than end-to-end autonomous operation. The figure outlines current AI use cases and core limitations across four representative sectors:

• 🏭 Manufacturing: Defect detection, engineering & design assistance, operational decision support
• 🌾 Agriculture: Plant disease diagnosis, agricultural decision support, consultation cross-validation
• 🏠 Real Estate: Property price estimation, transaction information extraction, property marketing copywriting
• 🏛️ Government & Public Administration: Administrative business consultation, social stimulation, administrative document writing

Growing industries mark the mid-stage of AI adoption, where Agentic AI evolves from a reactive assistant to an adaptive coordinator with embedded execution capabilities, operating within clear accountability boundaries.

In this phase, AI achieves deeper penetration, delivering dual-mechanism gains: enhanced operational efficiency and expanded business scope, while still requiring human oversight for critical decisions. The figure outlines the AI progression path, efficiency gains, and core challenges across four representative sectors:

• 🛍️ Trade: Document processing → compliance assistance → agentic workflow → business operation simulation
• 📰 Media: Structured generation → tool-using verification → multi-agent content production → towards multimedia integration
• 🎓 Education: Generic tutoring → pedagogical agents → multi-agent class simulation → evolving learning systems
• ⚖️ Law: General legal assistance → domain-specific case analysis → multi-step argumentation agent → multi-agent legal systems

Widespread industries represent the mature stage of AI adoption, where Agentic AI fully evolves into a self-governing production system with deep workflow integration, delivering dual-mechanism productivity gains under governed autonomy.

In this phase, AI achieves pervasive penetration, enabling end-to-end autonomous operation across complex, multi-step workflows while maintaining regulatory and ethical accountability. The figure outlines cutting-edge AI applications and expansion mechanisms across three representative sectors:

• 💻 Information Technology: Auto coding systems, iterative debugging, repository-level engineering
• 💰 Finance: Investment and trading systems, banking operations, insurance processing
• 🏥 Medical: Clinical systems, pharma R&D, public health services

If you find this repository useful, please cite our paper:

@article{ai4x2026ai4productivity,
  title={AI for Productivity in the Age of Agentic AI},
  author={{The AI4X Project Team}},
  journal={Preprints},
  year={2026},
  doi={10.20944/preprints202608.1248.v1},
  url={https://www.preprints.org/manuscript/202608.1248/v1}
}

Contributions are welcome!

If you would like to improve this repository, you are welcome to:

• add missing papers,
• revise incorrect metadata,
• update paper links and venue information.

Please feel free to open an issue or submit a pull request!