On the subject of AI

I'll use "LLMs" and "AI" interchangeably throughout this section, they function the same in this context.

This website was finished in under half an hour because of AI. Due to constraints with the desktop version of this design, you cannot click links or use the mousewheel on this face of the cube because the script is capturing pointer events.

Unless you view the mobile version or drag this cube face off-angle that is. Only then will you have access to the locked bounty of blog links below!

I sat in a café while the initial version of this website came down over Codex. A true Melbourne web developer moment: sipping coffee over CSS boilerplate. I'm trying out different agents, I'm not primarily a web dev and this is the new candidate off the slop-docks.

I keep this cube face in to remind me: current day AI is not a perfectly infallible Magic 8-Ball. Sometimes it leaks. I'll get around to the cube face issue eventually. Since this is not a heavy production website, the flaw serves as a reminder. AI may be fast at "fancy autocomplete" tasks and the trade-off usually comes in the form of non-functional UI.

AI is both transformative and dangerous. It can and does bring down production environments. We've all seen hype before - multiple - tech-hype - cycles - repeating - and the truth appears eventually.

Personally I don't believe we should abandon the use of AI in software development, nor could we. Even if it does produce - ongoing - slop waves. The technology is maturing in its own way, and this pressure lifts the industry. Now we need people in tech to shape this future for the better.

The answer is somewhere in between avoidance and adoption. AGI won't be found in the current generation of LLMs, and even if it was, enterprise adoption would be slow to adapt. As fast as software and emerging technology can be, people and rusted on business operations are everything but. If you've read this famous article you'll also recognise that grifters are reported to operate in the area. Human-in-the-loop style development processes are likely going to replace some workflows, even if they were slower with tools in 2025.

Developers are at this intersection, a tectonic shift and the fault line in progress; they are both using the tool and writing the tool. It will be they who figure out the fastest ways, and likely they who integrate it into their workflows before everyone else. In the meantime we still have to hand-roll certain parts of our code, the parts that LLMs cannot statistically reproduce without an existing template.

Now if you'll excuse me, my coffee is finished and with the time I saved on writing boilerplate I now will dutifully plough back into debugging the output of an AI. Welcome to the future.

(Update 18/04/2026) I never did get around to this cube face.