TL;DR
Install llm, drop three small files into your dot‑files, and you get CTRL +ALT +l to send the visible screen, CTRL +ALT +L to select N lines of scroll‑back, CTRL +ALT +N to start a fresh chat. Replies appear in a right‑hand split running less, and each pane keeps its own independent conversation.

Why?

I spend a lot of time staring at logs and stack‑traces wondering “what on earth?” Instead of copy‑pasting chunks into the browser I wanted a single shortcut that

Over the past year I’ve been refining how I build small tools and websites with the help of large language models. Below is a quick overview of the workflow that has worked well for me.

1. Create a GitHub repository – everything starts with a fresh repo so I can track progress and iterate quickly.
2. Clone locally – once the repo exists I pull it down to my development machine.
3. Kick off the OpenAI Codex CLI – for single-use tools I’ll start the Codex CLI right away. For larger projects, I write a high level README and include any example code so Codex has context. If the project needs an initialization step, I run that first so the LLM sees some starter code.
4. Iterate on the code base – I bounce between different tools and models depending on what I’m building. This lets me validate the output or add focused features without getting stuck in one workflow.
5. Push to GitHub – once the code reaches a decent state I push the changes. From there I use OpenAI Codex on the web to keep iterating. It’s great not having to be glued to my dev environment; I can request fixes and new features from anywhere.

Recently I’ve been experimenting with voice commands to drive the whole workflow hands-free. The approach is still new to me, but it has been fun seeing how natural language prompts translate directly to code changes. You can see some examples in these repositories:

For years, my blog sat dormant, a relic of a different era of the internet. The content was still relevant to me, a snapshot of my thoughts and experiences in the tech world from over a decade ago. However, the format was outdated, a mix of inconsistent markdown, old HTML elements, and front matter that didn’t quite fit modern static site generators like Hugo, which I’ve since adopted.

Manually going through dozens of posts, fixing front matter, converting HTML links and entities, and ensuring consistent markdown syntax felt like a daunting and tedious task. I needed a way to automate this, and that’s where AI came in. I decided to try aider, a command-line tool that uses AI models to help with coding and editing tasks.

So the latest version of the Windows 10 tech preview is out and it is pretty snazzy! Being the daredevil rouge that I am I decided to try it out on my work laptop (being the only Windows PC I have) and have had great success. While I wanted to try it out I didn’t want to chance destroying all my data and I also wanted an easy back out if I didn’t like it. Neither a format/reinstall nor an upgrade suit these requirements so it was time to try something even cooler, booting from VHDX!

Sorry all for the lack of posts, I’ve been lazybusy. But as you can see the site has had a new lick of paint but that is a whole different series posts. Instead I want to talk about my new NAS. Being the data conscious person that I am there is only one acceptable filesystem to use, ZFS. Thankfully there is also a nice appliance-like OS called FreeNAS that is based on FreeBSD and has ZFS as the foundation. Once I’d selected the OS it was time to get some hardware. By keeping an eye out I was able to pick up some awesome deals on Dell hardware (which I’ve listed below). With the hardware all installed and FreeNAS loaded onto a USB key I was finally able set up my system and have a look around.