Using AI Un-Sloppily

There’s a lot of talk about “AI slop” these days, and it makes sense. Increasingly, people are opting to offload their thinking to machines to the detriment of their intellect, and many companies are looking for any excuse to wedge bad AI features into their products to the detriment of their functionality. This makes it very easy and tempting to disavow generative AI entirely. However, I think it’s important to be pragmatic.

I, like many others, think of our current situation as being eerily similar to the dot-com boom of the early 2000s: revolutionary new technology leads to a stock market bubble, which leads to clumsy adoption of it by everyone and their dog until the bubble violently bursts. This parallel has been one of the main talking points against AI, and that’s fair. But people neglect an important detail: we’re living in the post-dot-com world, yet we’re still living with the Internet. I believe that, when the dust settles, after fortunes have been made and lost, we’ll still be living with generative AI. I simply hope that, through this, we’ll learn how to use it well, and use it sparingly.

There are many things that AI is bad at. There are also many things that it can do reasonably well, but that deterministic¹ tools can do better and more cost-effectively. It’s important that we don’t give into the hype and wedge AI into these areas where it doesn’t belong. But there are things it does well, too. I use AI sparingly. I would never use it to write my blog, mediate my communication with others, or do something that a deterministic program could do better, but I do still use it. I use it for boilerplate: for things that are mundane, which I know how to do but which are annoying and time consuming, in order to free myself up to do actual problem solving.

As I said, I don’t use it to write my blog, but when I decided to migrate my blog from raw HTML, with my own haphazard tagging and dating system, to Hugo, a Markdown-based static site generator (SSG), I immediately recognized that this was an application that made sense. I pulled down my old site’s source with SFTP, initialized a blank Hugo site, and told Claude Code² to have at it. It took some coaxing along the way, but it did a decent job. Rather than just copying and translating the raw HTML elements, I told it to rearrange things as it saw fit so that everything fit into the new template as naturally as possible. It didn’t change any of my posts (yes, I checked), and it even added tags (although I did have to clean them up afterwards). I even had it create a CI/CD pipeline to automatically build and deploy the site when I push changes to its source, which lives on GitHub. I spent about $7 in tokens for the whole ordeal, and in exchange I did in one hour (my lunch break, actually) what would’ve taken me the better part of a day. Furthermore, I don’t consider “fighting with my website’s SSG instead of writing content” to be a creative endeavour, and so I don’t feel like I missed out on anything.

All this to say: AI has worthwhile applications. Of course, I am vehemently against offloading your thinking to a machine, and am deeply concerned by the (rising) number people who do this. I’m also against wedging it into awkward places for its own sake. But there are places it shines, and we should acknowledge them, so we can use it in those areas to free the interesting problems up for us humans.