Today I had the need to convert a bunch of HTML into markdown. I searched the Apple app store for a Safari extension (yes I'm a Safari weirdo hold-out) and found one for $5, but it was from a developer I'd never heard of. I suspected the quality wouldn't be that great, and there was no linked source repository, and past experience tells me that it probably sucks and I'd just be throwing the money away.
So what are my other options? I remembered pandoc and some other CLI tools, but then the idea of pointing them at URLs, or saving pages to HTML and then converting them one by one sounded tedious. What I really wanted was a browser extension so I could just click "save" and be done.
I sighed and thought "eh, I guess I could just have Claude write a browser extension that just wraps pandoc for me real quick". The task raced through my mind. It would be so easy. It'd probably be wrong in a few ways, and then I'd have to screw with it back and forth a bit, but it would probably work, and then I could have the thing I need and move onto my actual task.
What stopped me is the knowledge that it would feel gross for all the reasons we all know that LLMs are gross. So I sat there for a few minutes just not moving forward.
And then I remembered that Obsidian has a little Safari extension who's purpose is to save web pages into your notes, and I even had the thing already installed – but I'd just never used it. So I added it's button to my browser, clicked on its settings, and right fucking there was a "Copy to Clipboard" button, and what it copies is Markdown!
This little moment is such a good example of how LLMs operate on our decisions making processes in such a negative way.
In the software world why have the DRY principle to guard against unnecessary duplication. The beauty of DRY is that we can invest in high quality abstractions around problems rather than re-inventing the same thing over and over again. It comes with a cost, of course (I've seen it taken to far more times than I can count), and we see a counter balance to those costs with the WET (Write Everything Twice) principle and various others which bring some nuance to how we balance these trade-offs.
LLMs, though, actively encourage a complete disengagement with this line of thought. It's just so easy to ask for a thing and get something in return. It looks basically correct? Right? And it was so easy? And in the age of the extreme degradation of search engines, it's getting so hard even discover if a solution already exists.
So instead of Don't Repeat Yourself (DRY) we end up with Constantly Repeat Yourself (CRY). And because what we're extruding at breakneck speed hasn't evolved and matured through trial and error, it's likely inferior to what it's replacing. Instead of building and accruing things of high quality, we're building disposable trash that we'll never maintain.
And it almost got me today, and it makes me sad because it's just such an astronomical waste of resources, time, energy, and effort.
