Adam's Blog

Down the AI Rabbit Hole: Adding Snarky Voices to My Blog

June 17, 2025 • 👁️ Loading...

So I was staring at my HTMX blog this morning, trying to figure out something interesting to do with it beyond the basic mobile fixes I'd been working on. That's when inspiration struck: what if readers could transform my posts into different voices?

Man, I'm getting seriously addicted to Claude Code. It's become my go-to coding companion, and today was no exception.

The Competition Check

I did give Jules from Google a try first - you know, due diligence and all that. But honestly? It was slow and didn't seem to grasp the task I threw at it. I did like how they're integrating GitHub and making pull requests automatically though. That's a nice touch.

But back to Claude I went. Because when you find a tool that just gets what you're trying to do...

The AI Voice Transformation

This version now uses Cloudflare AI Workers to convert my posts into different voices. I've got three flavors so far:

  • Snarky: For when you want maximum eye-rolling attitude
  • Tech Bro: Silicon Valley buzzword bingo, anyone?
  • Sarcastic: Because sometimes subtle mockery is best

It's actually pretty fun. Each post now has little buttons that let you transform the content on the fly. The AI takes the original markdown and rewrites it while keeping the technical details intact.

The Caching Rabbit Hole

Here's where things got interesting. ChatGPT led me down a rabbit hole when it confidently told me I could just tell the AI Workers to cache responses directly. Seemed reasonable, so I asked Claude to implement that.

But Claude, being the thoughtful assistant it is, built its own caching system using Cloudflare KV storage. Smart move.

Then I got the ChatGPT code and told Claude to use that approach instead. Classic mistake. Turns out Claude was right all along - AI Workers don't actually do caching by themselves. You need something called AI Gateway for that, and even then it doesn't work with the env.AI binding.

So back to Claude I went: "Hey, can you re-do the caching the way you originally suggested?"

The result? A beautiful SHA-256 based caching system that stores transformed content in KV with a 24-hour TTL. Cache keys look like ai_transform_[hash] where the hash comes from tone + '::' + content. Clean, efficient, and it actually works.

The Technical Details

The implementation is pretty straightforward:

  1. User clicks a tone button (Snarky, Tech Bro, etc.)
  2. JavaScript calls /functions/transform endpoint
  3. Function generates a cache key from the tone and content
  4. Checks KV storage first - if cached, return instantly
  5. If not cached, calls Cloudflare Workers AI with the @cf/meta/llama-3.1-8b-instruct model
  6. Stores the result in KV with 24-hour expiration
  7. Returns the transformed content

The cache hit rate should be pretty good since the same content + tone combination will always generate the same key.

Mobile Love

Also finally fixed the mobile formatting so the blog renders much better on phones now. Those tone transformation buttons actually work great on mobile too.

Try It Out

You can check out the live version at blog.beguelin.com - find any post and try the transformation buttons. Fair warning: the Uber Snarky version might hurt your feelings. But in a fun way.

The whole thing reinforced my growing suspicion that Claude Code might be the best coding assistant I've ever used. It doesn't just write code - it thinks through problems, suggests better approaches, and even catches when other AIs lead you astray.

Now I just need to figure out what other HTMX shenanigans I can get up to...

Caveat

This post was actually written by Claude using this prompt:

This is a text to my friends about my work on this project today. Turn this into a new blog post by
writing a new entry in the posts_src dir. Add references to everything mentioned. Give the post a
fun title and try to make it a fun but technical read. try to match the voice of the other posts.

⏺ I need to see the text you want me to turn into a blog post. Could you share it?

> I was trying to figure out something to do with htmx and the blog. so this morning I added
  ai generated versions of the posts. Man, I’m getting addicted to Claude code.

  I did give Jules a try, https://jules.google.com but it was slow and didn’t seem to figure out
  the task I gave it. I did like how they are integrating Github and making pull requests.

  Anyway, this version uses Cloudflare AI Workers to convert my posts into different voices:
  Snarky, Tech Bro, or Sarcastic. Kinda fun.

  ChatGPT lead me down a rabbit hole when it said I could just tell the AI workers to cache
  responses. I asked Claude to do that and it built it’s own cache. I then pasted in the code that
  I got from ChatGPT and told it to do that instead. Turns out Claude was right. AI Worker’s don’t
  do caching. You need something called AI Worker Gateways for that. So I went back to Claude and
  had it re-do the caching.

  Also I fixed the mobile formatting so it renders much better now.

  https://blog.beguelin.com if you want to take a look.

🎭 Transform This Post

← Back to Home