Easter Weekend AI Project

Sat, 04 Apr 2026 00:00:00 +1100

An Easter Weekend Experiment with Copilot CLI

Over the Easter weekend I decided to get my hands dirty with something a little different. No production workloads, no architectural diagrams, no pressure. Just a small, contained experiment to see what GitHub Copilot CLI actually feels like when you let it roam a bit.

The target was my personal resume site. It already existed and technically did the job, but if I’m being honest it was being held together with good intentions and duct tape. Functional, but inconsistent, messy, and very clearly something I’d been meaning to clean up “one day”.

This felt like the right kind of project to experiment on.

The Mission (Intentionally Vague)

The brief I gave Copilot CLI was deliberately loose: modernise it.
No wireframes. No detailed spec. No careful breakdown of tasks. I just pointed it at the repo and let it ask questions.

That looseness was intentional. I wasn’t trying to see how well it followed instructions, but how it behaved when it had room to propose rather than just execute.

From Idea to Running Locally

After a bit of back‑and‑forth and a couple of nudges from me on concepts I cared about, Copilot CLI put together a plan, created a new branch, scaffolded the changes, handled npm, and got a local version running.

End to end, including me stopping to take notes for this post, the whole thing took around 30 minutes.

That’s not nothing.

What Actually Changed

Cosmetically, the result isn’t a dramatic departure. If you didn’t know to look, you might not even notice much difference at first glance.

Under the hood though, it’s a different story. The code is cleaner, better structured, and actually commented throughout in a way the original never was. The site feels like something I’d be comfortable picking up again in six months without immediately regretting past‑me’s decisions.

There were also a few small additions that came from giving the AI a bit of creative freedom. A simple theme switcher, for example. Slightly weird, not strictly necessary, and probably something I wouldn’t have bothered implementing manually—but I was happy to have it along for the ride.

I’m not a UX designer, and I didn’t pretend to be. But letting Copilot suggest alternate approaches to layout and UX, rather than just fixing exactly what I pointed at, genuinely lifted the site in ways I wouldn’t have thought to ask for.

Why This Was Interesting

The most useful part of the exercise wasn’t speed, and it wasn’t raw code generation.

It was what happened when I stopped treating the tool like a very fast typist and instead let it propose options. Plans. Alternatives. Things to react to rather than instructions to follow.

That shift—from “do this” to “show me what you’d do”—was where the value showed up.

A Thought Before You Try Something Similar

This kind of experiment works precisely because the stakes are low. I wouldn’t point Copilot CLI at a regulated production system and say “modernise it” with a straight face.

But if your AI usage still looks like autocomplete, boilerplate, or shaving a few minutes off typing, you’re probably missing the more interesting part.

If you’re going to play with these tools, do it somewhere safe—and do it in a way that forces you to exercise judgment, not just measure speed.

That’s where things actually start to get interesting.

AI Weekend Projects on Brewed in the Cloud by Chris Hailes