Generative AI

This course encourages the use of Generative AI (GenAI) tools such as GitHub Copilot, ChatGPT, and Claude Code when developing your project. However, the usage of these tools can greatly impact our expectations for your project. We all understand how powerful they can be, but also how they can lead to projects that are poorly understood by the developer and code that is not easily maintained. This page will attempt to clarify our approach to handling project expectations in light of GenAI.

To begin, we wish to impart a core mindset for grading projects in this course:

To aid in the discussion, we’ll loosely group projects into three categories. Those categories are not mutually exclusive, and some projects may not fit neatly in a single category. They are simply useful benchmarks to start from.

No GenAI Usage

Students may choose to build a project that does not involve any GenAI usage when it comes to developing the actual code of the project. They may still make use of GenAI to create presentation graphics, diagrams, and other artifacts for the project. The intent of a project in this category is to give the student hands-on experience with building a project’s code directly from scratch, just like it had been done for decades prior to the release of modern GenAI tools.

As such, our expectations for this type of project will reflect what we expect a single developer to be able to achieve independently. Thankfully, most faculty and advisors are very familiar with these types of projects, and have a solid idea of what a reasonable scope and scale for such as project is.

Students who choose this type of project should be prepared to discuss their code in-depth and explain their thought processes, debugging strategy, lessons learned, and overall development process that is reflective of a project that does not use GenAI tools.

Violations: Students who declare that their projects do not make use of GenAI will be held to that requirement. Any violation of this policy may be treated as a violation of the K-State Honor Pledge and may result in a grade of 0 for the assignment and other sanctions approved through the K-State Honor Council.

Assistive GenAI Usage

Students may choose to build a project with the use of assistive GenAI tools such as GitHub Copilot’s autocomplete feature, or by using standalone chats with tools such as Claude and ChatGPT. More advanced tools such as Claude Code or full agentic use of GitHub Copilot is not included in this category.

Our expectations for this type of project are somewhat higher than a project that does not use GenAI at all. We expect that students may spend less time writing actual code and debugging, and more time developing the overall structure of the application and the user-facing features. Therefore, we hope to see a project that is more feature-complete and has more features overall than a project without GenAI usage.

Students who choose this project should still be prepared to discuss their code and structure in-depth, but should also discuss how they were able to make use of GenAI to assist in answering questions, writing code, and solving problems throughout the project. This discussion should be reflective of a student who used GenAI in an assistive capacity, but did not make use of any agentic features.

Violations: Students who declare that their projects do not make use of agentic GenAI will be held to that requirement. Any violation of this policy may be treated as a violation of the K-State Honor Pledge and may result in a grade of 0 for the assignment and other sanctions approved through the K-State Honor Council.

Agentic GenAI Usage

Students may also choose to make full use of an agentic GenAI tool such as Claude Code to develop their project. This category is meant to be reflective of what students may actually experience in industry, but also comes with significantly higher expectations and requirements. Agentic GenAI tools are very powerful and can lead to an extraordinary project, but may also cause developers to lose sight of the structure and contents of their code, making it much more difficult to debug and explain.

Our expectations for this type of project are therefore very high - we expect the project to have a number of features that are complete, tested, and usable. Other items such as security, accessibility, and documentation may also be expected (many of those items are a simple command or skill that can be invoked).

Students using agentic GenAI should be prepared to describe their full setup, including key files such as CLAUDE.md, skills, plugins, and models used. This process involves many design decisions that should be considered with the same amount of care as any design decisions made in the code itself. In addition, we encourage students to have their agents create and update development plans that are stored in the repository alongside the code (most agents can be configured to store these plans). This helps make the process and context visible to the instructor and advisor.

That said, we still expect students to have a solid understanding of the structure and functionality of their projects, including the ability to step through the structure of the code and design documents.

Students who declare that they plan to use agentic GenAI may not be at risk of violating the K-State Honor Pledge, but instead assume the risk that the expectations for their project will be reflective of what we expect a student to accomplish with that level of assistance. If a project does not meet those expectations due to poor or lacking usage of GenAI, that will be reflected in the student’s overall grade on the project.