AI Usage Policy

As AI tools have become increasingly prevalent in content creation, I believe it’s important to be transparent about how I use them on this site and in my work.

My approach to AI is pragmatic: it’s a tool that amplifies human capability when used thoughtfully, not a replacement for human creativity, expertise, or judgment. I use it to handle tedious tasks, catch errors I might miss, and accelerate certain aspects of my workflow–but the creative direction, critical thinking, and final decisions are always mine.

Regular readers have come to expect that kind of thing from this site over two decades, and I intend to keep it that way.

So get over my use of em dashes already.

All written content on this site is human-authored. I do not use AI to generate articles, blog posts, or any other written material. Every word, sentence, and idea originates from me, reflecting my own experiences, research, and perspectives accumulated over 20+ years of writing.

This is a fundamental principle: the thoughts, opinions, and narratives here are mine, not algorithmically generated approximations of what a “tech blogger” might say.

While I don’t use AI for content creation, I do leverage it extensively for revision, proofreading, and quality control:

• Grammar and consistency checks: After finishing an initial draft, I use LLMs inside VS Code to catch grammatical errors, awkward phrasing, repeated content, and missing words (as I’ve written before several times, sometimes I write posts over multiple weeks, and inconsistencies can and do creep in)
• Link validation: This site is actually a wiki, and I wrote a wiki-helper MCP server specifically to help with that. It helps me verify wiki links, fix broken references, optimize images, etc.
• Formatting standardization: The same MCP server helps ensure consistent frontmatter, heading structure, and reference-style links across the site, as well as playing an increasing role in converting older posts to modern Markdown standards.

I’ve detailed this workflow in my post about per-project MCP servers and many other places, but the key thing to remember is that I am not getting any younger, and these tools help me make the best use of my time to maintain quality and consistency in my writing.

I have a background in print design and editorial work (which helped pay my way through college back in the desktop publishing era), where I’ve always valued the interplay between text and visuals.

Decades after I stopped using Photoshop, Illustrator, and PageMaker, I’m still interested in how images can enhance storytelling, and love spending hours tinkering with visual design–and, in the modern age, prompting the AI image generators du jour.

So I occasionally use AI-generated images on this site, primarily for mood setting and creative experimentation (I’ve found that images are an excellent side channel for conveying meaning to complement the written word, and sometimes spend as much time on image prompts as I do on the text itself).

Depending on what tool I used and whether or not it allows me to capture that information, I will add the prompt used to generate the image in the alt text for transparency.

Think of it, again, as a force multiplier for my own creativity, not a substitute for it.

I also use AI as a force multiplier for coding in general, but where it relates to this site, its entire build system and static site generator are built around a custom Python codebase that I’ve been evolving for over 15 years now.

Of late, I have been routinely enhancing and maintaining it with the help of AI tools, and working on a streamlined port that is intended to be faster and more modular.

In general, the things I typically use AI for in my development workflow include:

• Generating scaffolding and boilerplate code
• Writing unit tests and standardized logging
• Creating UI components that I then refine and polish
• Reviewing code for potential issues
• Exploring unfamiliar codebases and technologies

As a side note, I’ve found that having proper tests and linting in place significantly improves LLM performance, even considering that right now we are still at a point where I need to verify the tests themselves.

The main AI tools in my workflow are:

• A custom Obsidian plugin to clean up most of the original cruft from my notes
• GitHub Copilot integrated into VS Code and vim
• OpenAI models (including the Codex variants)
• Custom MCP servers for specialized tasks (built with my own MCP library)
• Various image generation models for experimental and illustrative work

I don’t currently block AI crawlers from indexing this site. My content is licensed under [CC BY-NC-SA 4.0][cc], which allows for sharing and adaptation with attribution, non-commercially, etc., so there’s a huge, abyssal grey area here that Cory Doctorow would have kittens over–but realistically there’s not much I can do about it anyway. It would be kind of fun to end up as part of humanity’s collective transhumanist digital unconscious, so I don’t usually bother unless someone tries to impersonate me or misrepresent my work.

If you have questions about how I use AI or want to know more about specific aspects of my workflow, feel free to get in touch.