A [Vibe] Code-Along

Vibe coding—a new and loosely defined term in software development that refers to the practice of prompting AI tools to generate code rather than writing code manually—has become more common in recent years as Anthropic’s and OpenAI’s models have grown more sophisticated and capable of producing functional code. A developer friend compared the process to having an intern or junior developer write code that a senior developer reviews and corrects. I have heard a similar analogy applied to legal research, with AI acting as the junior associate and the lawyer serving as the supervising partner who reviews the work. But that is a topic for another day.

For this post, I want to explore how tools such as Claude Code can help users learn the fundamentals of coding and apply those skills in creative ways.

Before diving in, I want to explain my approach to artificial intelligence. I am vehemently opposed to using AI to replace human labor or actions. I believe we must discuss AI as a tool, use it as a tool, and treat it as a tool—not as something more. I have no interest in replacing my writing with AI-generated slop or allowing AI to perform my research for me. Instead, I believe AI works best when integrated into an existing workflow. Much like using Word instead of paper or Google instead of an index, this technology, when used with care, can help me complete a task rather than complete the task for me.

With that philosophy in mind, I began experimenting with Claude Code to help me learn to code.

I took computer science courses during both my undergraduate and graduate studies, but I have struggled with writing code. I blame dyslexia, although learning to code is a difficult undertaking in its own right. I have stacks of books on learning R and Python, but learning programming from a book presents its own challenges, especially for someone like me who prefers hands-on learning.

Enter Claude Code.

Of course, this approach has a significant limitation: I am not a senior developer who can evaluate the quality of the code an AI system produces. Instead, I must rely on a basic question: “Does it work?”

I have found that discussing my projects with friends who understand programming and consulting online forums helps me identify mistakes and understand where the AI—or I—went wrong. As a learning environment, however, Claude, GitHub, and government documents have provided an excellent opportunity to experiment with Python commands and learn how to implement them.

Why government documents, you ask?

Because government websites are open, provide freely accessible information, and are notorious for their difficult navigation and search functions. This led me to a question: What if I could learn to code while developing tools that improve aspects of my job?

I began with what I thought would be a simple project. I wanted to create a system that would send me personalized emails identifying new docket entries in active Supreme Court cases without relying on Bloomberg or PACER. Since the Supreme Court’s website maintains active dockets that receive updates, I figured this would be a manageable first project.

And, for the most part, I was correct.

I opened Claude Code and prompted it to help me write the code while explaining each step of the process. Claude generated the code, explained the purpose and function of each line, and walked me through implementing the tool in a GitHub repository—something I had attempted only once before, a long, long time ago.

And it worked!

Mostly.

I had to reconfigure a few components, and I discovered some errors in my original design. The emails identify new docket entries but do not include case names or other important identifiers, something I will need to correct in a future iteration.

Nevertheless, the project was a successful first experiment. I learned how to approach the coding process, gained experience implementing a tool through GitHub, and created something that performed a useful function. More importantly, I found the process informative and fun.

This experience has encouraged me to explore other ways to apply coding solutions to government websites that are difficult to search and navigate.

Throughout my work as a legal researcher, I have encountered questions where I know a document exists and should be accessible, but the available search tools and navigation options are either unintuitive or nonexistent. These situations can turn what should be a straightforward search into a frustrating research project.

I hope that combining my legal research expertise with my developing coding skills will allow me to approach these problems in new ways. By learning to create tools that address specific research challenges, I may be able to expand what I can accomplish as a legal researcher.

Whether these experiments lead to new solutions or more time spent debugging code remains to be seen.

But only time will tell.