Claude-Mini: Complete Resource Pack for the 4-Hour Agent AI Academic Research Workshop at the University of Notre Dame

Professors from the University of Notre Dame's Business School have open-sourced the Claude-Mini project, providing complete teaching resources for a 4-hour Agent AI academic research workshop to instructors in economics, finance, and accounting. These resources include Quarto slides, six live demonstration packages, and supporting practice materials.

With the rapid development of large language model technology, the academic research field is undergoing a profound shift from traditional tool usage to agent collaboration. For researchers in social science fields such as economics, finance, and accounting, how to effectively use AI agents to improve research efficiency, enhance paper quality, and accelerate replication processes has become an urgent practical issue. However, most academic training resources are either too technical or too general, lacking systematic guidance tailored to specific disciplines.

Professor Pedro Sant'Anna from the University of Notre Dame's Mendoza College of Business敏锐ly captured this need and launched the Claude-Mini project—a complete resource pack for a 4-hour Agent AI academic research workshop designed specifically for business school instructors. This project not only provides carefully designed teaching content but also uses a "watch-along" training model, allowing participants to intuitively observe how experts collaborate with AI agents to complete actual research tasks.

Claude-Mini is a teaching distillation version of the claude-code-my-workflow project, not a simple branch. It is specifically aimed at business school instructors in economics, finance, and accounting, with the goal of helping participants establish a systematic understanding of the application of agent AI in academic research through a half-day intensive training.

Unlike traditional hands-on training, Claude-Mini adopts a "watch-along" mode: Pedro conducts real-time demonstrations on-site, and participants observe and learn without needing to install any software or configure the environment before the training. This model lowers the participation threshold, allowing instructors to focus on understanding the working logic and thinking patterns of AI agents rather than being distracted by technical details.

The entire 4-hour workshop (including two 10-minute breaks, totaling 4 hours and 20 minutes) is carefully designed into seven sections, each with clear teaching objectives and time allocation:

This section lays the conceptual foundation for the entire workshop. Participants will learn the essential differences between agent AI and traditional AI tools, as well as how to establish collaboration norms with AI agents through CLAUDE.md files. This section includes micro-interaction demonstrations showing how to set the agent's working context.

This is the first core demonstration session of the workshop. Pedro will show on-site how to use Claude Code to improve a specific section of an academic paper from start to finish. Participants will observe how the agent understands academic writing norms, how it proposes constructive revision suggestions, and how it engages in iterative dialogue with researchers to achieve the final goal.

This section focuses on skill design—how to encapsulate repetitive research tasks into reusable AI skills. Participants will learn to design custom skills suitable for their own research workflows through paper exercises. This session emphasizes practical thinking rather than on-site coding.

As AI agents' capabilities improve, multi-agent collaboration has become a new research paradigm. This section introduces how to orchestrate multiple professional agents to collaboratively complete complex research tasks, and includes a mini-demonstration showing the practical application of multi-agent systems.