I’ve been (happily) using Quotion to run this blog for the last two years. At the time, I was very focussed on finding ways to simplify my life and the Wordpress blog I was using felt anything but.
Quotion was everything I needed and nothing I don’t. Write a note in Apple notes and it goes live. It’s a brilliant tool and you should definitely try it - irrespective of my move away from the tool, I think it’s brilliant.
Over time though I felt a number of small things nagging at the edges of my brain. The spacing that wasn’t just so or the way the titles rendered on post pages. There are ways to fix many of these issues in Quotion using custom css, but with every fix and edit, if felt like I was moving back towards complexity. Back towards “things to keep track of”.
The one that did the camel at last was image sizes. Just out of interest I ran a page speed test on my side. The results were not great. Images were massive, not optimised and trashing my FCP. And there was absolutely nothing I could do about it.
This lack of control was what sent me chasing after a new tool. I interviewed Hugo (too finicky) assessed Ghost (too powerful) and even thought about going back to Wordpress(!). Nothing was <u>really</u> what I wanted.
Eventually, while building a new personal relationship management tool for myself (Marbles) I thought - what the hell, why not build the publishing platform I actually want?
And thus BoCin was born (pronounced like boatswain, a word I love because of how diabolical it is).
Bocin is essentially a blog built on Astro, but with the same underlying functionality that I love about Quotion.
I write a note, BoCin goes to fetch it, turns it into a blog post and then puts it on my site. As I sit here right now, all of this is happening in Apple Notes. Once I’m done, I’ll simply remove the draft section from the front matter and within about 15 min or so this post will be live on my site.
The very definition of simplicity.
Building a blogging platform with an AI coding assistant
I can’t code. Nothing beyond some basic HTML and CSS really. So in order to build my own tools I (like the rest of the world) has been using a combination of AI coding tools to help me ‘build with prose’. Sometimes it’s with Lovable, sometimes Qwen running locally in Ollama and sometimes, as in this case, it’s Codex.
As is my usual process, I spent the bulk of my time chatting about HOW the tool would work and what it would it do long before I asked Codex to write the first line of code. In this case at least the answer to the most important question - who is this for - was easy: Me!
As for the problem statement, that was also easy to define. So after getting the spec MD document in good shape I set off asking the model to build me mockups of screens that I would need: settings, admin, analytics tracking etc. We refined over time and then got the real deal going. Within about 2 days, I had the admin up and running (although only locally).
The scary part - giving AI computer use authorisation
Working with the Computer Use function on ChatGPT was the scariest part of this build. It’s an eerie feeling to see the mouse dart across the screen knowing that there is a pseudo-sentient thing controlling it in service of my app build.
If I’m honest, it took me a good long time before I felt comfortable giving this access. I even had to make a new machine profile for ChatGPT, just to make sure the machine stays in its lane and doesn’t get in to all of my personal stuff.
The boring parts
Like any project, there’s more boring stuff to solve than there are excited moments. Here are but a few:
- AWS configs. So. Many. Settings.
- Figuring out why the app ballooned to 19GB in size (bad devops practices)
- Figuring out why the git repos were not working correctly
- Debugging publication issues (and debugging AWS not accepting publication)
- Figuring out why front matter wasn’t working in some cases
- Learning about IAM policies
- Reading up on good software backup policies
And the list goes on. Overall I like that it extends me a little bit out of my comfort zone and helps me get exposed to things I might otherwise never encounter.
Software is still hard
While Codex did build me a working tool, it had several problems. Because it didn’t specify how to handle errors, an AWS publication error triggered an endless loop, with each loop creating a new build package. Eventually the total number of packages reached 19GB and my Mac mini gave up the ghost. These are very basic things that an experienced developer would 1) know not to do and 2) probably notice before it killed the host machine.
Tools like Codex look and feel like magic but sometimes the illusion doesn’t hold up to close inspection. I foresee a lot of troubleshooting and debugging in my future, including getting frustrated with the decisions that the tools make.
For now, I’m having fun building BoCin and I’m sure I’ll keep plugging away at it for a few more months before I’m 100% happy with it. I do like that it’s mine and that I can make it work just so, just for me. It gives me a certain degree of satisfaction knowing that I’m working with a bespoke tool, even if it just to write down what amounts to brain doodles that only 2 people on the internet will read.
This blog post is 100% human written by me, Lester Hein, which is why it’s probably full of mistakes, errors and other things that a machine would not do. Take that as you will.
