I prefer to write most code myself, but due to the growing prevalence of AI in software development, I am increasing my use of it to stay current on the technology. Code that is heavily or exclusively AI-generated ("vibe coded") will include a disclosure header indicating as much. The vast majority of what is published under https://github.com/ericpruitt/ will continue to be hand-written, and I have created the slopvat organization to host vibe coded projects. All writing that is not part of AI-generated code is written by myself with AI used only for proofreading and editing; I write my own blog posts, emails, and documentation — as long as I also wrote the code. This currently applies to both my personal and professional writing, but that could change for the latter if I am compelled to use LLMs for writing by my employer.
The AI-generated code that exists under my main account generally falls into a few categories:
- Scenarios where I see coding as a means to an end, and I don't see a lot of value in spending time figuring out the problem on my own. See my post "LLM Laundering for Security" for one set of examples of this.
- Implementing complex algorithms or mathematical calculations, especially those that come from research papers where, even if I understand the logic, there is a high risk of transcription and/or translation errors.
- Code published by other people that is incorporated into my projects which may have been written by LLMs.