It's funny that "restructure the codebase to be more friendly to agents" aligns really well with what we have "supposed" to have been doing already, but many teams slack on: quality tests that are easy to run, and great documentation. Context and verifiability.
The easier your codebase is to hack on for a human, the easier it is for an LLM generally.
It’s really interesting. It suggests that intelligence is intelligence, and the electronic kind also needs the same kinds of organization that humans do to quickly make sense of code and modify it without breaking something else.
I had this epiphany a few weeks ago, I'm glad to see others agreeing. Eventually most models will handle large enough context windows where this will sadly not matter as much, but it would be nice for the industry to still do everything to make better looking code that humans can see and appreciate.
This reminds me of the early days of the Internet. Lots of hype around something that was clearly globally transformation, but most people weren't benefiting hugely from it in the first few years.
It might have replaced sending a letter with an email. But now people get their groceries from it, hail rides, an even track their dogs or luggage with it.
Too many companies have been to focused on acting like AI 'features' have made their products better, when most of them haven't yet. I'm looking at Microsoft and Office especially. But tools like Claude Code, Codex CLI, and Github Copilot CLI have shown that LLMs can do incredible things in the right applications.
I’ve thought about doing something similar, but at the Service Worker layer so the page stays the same and all HTTP requests are intercepted.
Similar to the window.stop() approach, requests would truncate the main HTML file while the rest of that request would be the assets blob that the service worker would then serve up.
The service worker file could be a dataURI to keep this in one file.
I think there’s an uncanny valley effect with writing now.
Yesterday I left a code review comment that someone asked if AI wrote it. The investigation and reasoning were 100% me. I spent over an hour chasing a nuanced timezone/DST edge case, iterating until I was sure the explanation was correct. I did use Codex CLI along the way, but as a power tool, not a ghostwriter.
The comment was good, but it was also “too polished” in a way that felt inorganic. If you know a domain well (code, art, etc.), you start to notice the tells even when the output is high quality.
Now I’m trying to keep my writing conspicuously human, even when a tool can phrase it perfectly. If it doesn’t feel human, it triggers the whole ai;dr reaction.
They changed some words pretty much right after the acquisition. There was some controversy when they started doing "themed" words (like Christmas stuff in December) vs more "random" words. Some words were also removed for having negative vibes/political liability
They removed WENCH from the list of upcoming solutions fairly quickly, but forgot to add it back to the list of available words so you couldn't use it as a guess for a little while. It made it back to the list eventually.
I believe these lists are more like what is described in the blog post. Diction of words, filtered to 5 letter words, no plurals, etc. It most likely has 99%+ of the words, but maybe some they don't actually use in Wordle.
I've used my own tool (https://pseudosavant.github.io/ps-web-tools/wordle-solver/) for understanding how many words are left after each guess. It'll show hints if you want them too, but they are disabled by default. I like understanding how my guesses reduce the word space well (or not).
It uses the list of all of the words that can be in Wordle, and there are so many words I can't imagine anyone guessing. And I come from a family with large vocabularies.
The easier your codebase is to hack on for a human, the easier it is for an LLM generally.
reply