My current use of AI

I see a lot of AI slop online these days, including in blog posts, and I am confident that many words today are not written by humans. I do not mind the use of generative AI to write per se. My issue is the lack of transparency about the use of AI. For example, I stopped reading anything from Tyler Cowen when I found out that he used AI to fabricate quotes on his blog without being transparent about this fact. People can use AI as much as they want to keep the content production game strong, but why try to hide the use of AI? (Especially if you believe AI is going to change the world.)

I do not like the current trend where people feel the need to declare in individual posts that it is not written by AI (as an example, I was reading this post that explicitly states "no LLMs were used in the writing"). Should I assume everything else written in your name is actually written by AI? When I go to a restaurant, I do not want the waiter to tell me that today the food is prepared in the kitchen. I do not care about what a person did not do - only what they did do. The default should be that if you publish anything in your name, it is written by you and not somebody else (including AI). I am aware that defaults and norms can change, but I would prefer to see people being transparent and explicit about their use of AI.

Here I will outline how I currently use AI. While I try to stay in the loop with new developments, I am relatively slow to change my workflow, so do not expect any state-of-the-art insights on how to use AI. If anything, the takeaway message is that I do not have a unified approach on how to use AI, and particularly not a 'new is great' mindset (despite knowing that the most recent models, on average, are better than older models). If you care about trends and new tools, you can get by by following a small number of people such as Simon Willison (for technical updates) and Ethan Mollick (for applied updates). As a rule of thumb, if anything is worth paying attention to, they will eventually cover it.

That being said, I have certain principles I comply with, and I do not expect to update those in the future. I do not use generative AI in any shape or form to write any content on this blog. Not a single word is written by anybody but myself. This is an active choice as I believe generative AI can write more and in some cases better content than me. However, the raison d'être of my blog is not to publish multiple posts a day with content that might as well come out of a good prompt provided to your favourite LLM. If anything, the purpose of this blog is not to "create content". I write to share my words. So, unless otherwise stated, none of the words you will find here, in the past or the future, will be written by AI. Similarly, I have never used or will let AI write anything in my name (books, scientific papers, articles, etc.).

I do use AI to read my blog posts. I write my posts in Visual Studio Code in .md files without any autocorrect or spellchecker or the like, and I find it useful to use the ask mode in GitHub Copilot to check if there are any typos in what I have written (there usually are!). I manually edit the text and fix any typos. I do not use the edit mode or agent mode to make any changes to my posts. This might be easier, but I do not feel comfortable letting AI automatically edit any of my words. Accordingly, I am not keeping AI completely out of my workflow for my blog, but I use it to read and comment on my writing, not to write or make any changes. I could experiment more with AI giving detailed feedback on my posts beyond identifying typos, but this is not something I have explored.

For my work on different projects, I rely on a combination of ChatGPT (Classic and Codex) and GitHub Copilot. I find working with agents extremely useful, and the quality of the work is in most cases better than what I can do on my own without AI, especially when taking the time component into account. I usually open a project folder in Codex, let the agents cook, and then manually work with and check the relevant files in Visual Studio Code and make smaller edits with GitHub Copilot if needed. There is a Codex integration for VS Code, but I do like to separate parts of my workflow and use version control (git) to check edits. It is a great feeling to see agents working on different projects, and while it is not necessarily the feeling of working, it is a feeling of being productive.

For some projects I do not write any words or code myself beyond instructions, and I like it that way. There is simply no reason for me to waste time on writing code that can be done more effectively and better by AI on specific projects. It is a lot easier to get things to work than it was when I had to copy-paste code snippets from Stack Overflow. I am neither vibe coding nor tokenmaxxing, as I do like to follow along and understand the code and be able to make manual edits if I so desire. For some projects, like this blog, I care about the words and the code being my own, and for the projects where this is not a priority, I make sure to document the use of AI (i.e., save the prompts and any other details that can be important for people, mostly my future self, to reproduce). Most of the work I do today is not publicly available, but should I at any point in the future make any work public where AI is doing any heavy lifting, I will of course be transparent about this.

I have a feeling that my exact workflow will be outdated soon. I mean, not the fact that I will never let AI write my blog, but the use of particular tools and their abilities. Things are changing a lot these days, weeks, and months, and I am curious how I will look back at my current use of AI in the future. It has been less than five years since ChatGPT with GPT-3.5 was launched but it feels like ten years ago since I last consulted Stack Overflow for help with a programming task.

The interesting thing is that it is not possible to not use AI in 2026. The moment you make a Google search, chances are you are engaging with an AI assistant that helps create content. As the norms around the use of AI change, not only might we not reflect upon how we use AI to consume content, we can begin to not reflect upon how we use AI to create content. I am seeing more people using AI without thinking about how they use AI, and we are slowly beginning to not think of "the use of AI" as AI will be an integrated and natural part of how we use different tools (I made a similar point in a post back in 2024). That is, we might not think about using AI as something separate from other tools, and this blog post might be equivalent to having written my thoughts down in the 90s on "My current use of the Internet".

My current use of AI is relatively boring, and I like it that way. I prefer to be very conscious of what parts of my workflow are being affected by AI and why. In the future I hope that we will see additional progress on local models, and especially the ability to use agents without having to share a lot of data. That is, I hope my future use of AI will be less dependent on a few tech companies in the US, but maybe that is too much of an ask. To be continued.