So, one Sunday evening I thought I’d give the whole “vibe coding” thing a proper blast.
I wanted to test what Copilot Pro, using the ChatGPT 5.5 Think Deeper model, was actually capable of when pushed beyond the usual “rewrite this text” type of request. Rather than doing something sensible, I thought, why not make a Solitaire game for my blog?
So here it is, Sweet as Solitaire (more about it below)…
What started as a quick experiment turned into something much more involved. I did get a bit carried away, but that was kind of the point. It turns out this kind of back-and-forth coding is quite addictive, especially when you can test something, spot what is broken, explain the problem, and have the code updated almost immediately.
The interesting bit for me was not just that it could make a card game. It was how far it could go with the game logic.
The hint system (which came about when building the ‘no more moves’ function), ended up being far more complex than I originally expected. If you bring up the hints, especially the “Show hints for this game” option, you’ll get a sense of the deeper computation going on behind the scenes. The game is not just looking for obvious moves, it is doing a kind of lookahead, testing possible future moves, and trying to work out whether there are still meaningful paths forward. That was built mostly by playing the game, finding edge cases, and talking through the logic step by step.
Some of the styling still needed manual CSS edits, especially where I wanted things to sit cleanly in the page or work properly inside WordPress. But a lot of the visual themes, including the more playful Retro Future inspired design, were generated through prompts and then refined through a mix of testing and small manual fixes.
Is it perfect? No.
There was a lot of back and forth. Sometimes it would solve a genuinely tricky problem, then unexpectedly drop a feature that had already been working. Other times it would edit some unrelated part of the code and break something that had nothing to do with the change I’d asked for. That became one of the main lessons from the process.
If you are using AI to help write code, it pays to be very specific. I found myself repeatedly saying things like:
Make this change, but ensure all other functions remain unchanged.
Or:
Compare this version against the original working version and summarise exactly what changed.
That kind of instruction matters. It also pays to ask for a change summary, then actually check it. The AI can move quickly, but it can also be very confident while quietly breaking something. So the workflow becomes less like handing over the job completely, and more like working with a fast, slightly unpredictable assistant.
Still, the result is pretty fun.
The game includes scoring, undo, draw-one and draw-three modes, strict Klondike rules, multiple visual themes, high scores (for user only, no global high score [yet] but they are saved as a cookie in your browser), no more moves and hint logic, best times, and a few Kiwi-flavoured messages along the way. It also works as a single HTML file, which I still find impressive. No build process, no server-side code, no complicated install, just one self-contained file running in the browser.
That might not be revolutionary, but it is a nice reminder of how much can still be done with plain old HTML, CSS, and JavaScript, especially when you have an AI assistant helping push things along.
You can also play it here on its own page.
It is called Sweet As Solitaire, with the very accurate tagline: Sweet as when it works, yeah nah when it doesn’t.