Skip to content

Adventures in AI: How I Accidentally ‘Vibe Coded’ a Video Game

It’s fair to say that my planned AI-in-healthcare career reboot has taken a bit of an unexpected turn.

After resigning from my healthcare management post in October, it was fully my intention after a short break to then step back into the healthcare fray. That endeavour began with some fevered absorption of online courses in AI and Agile development.

However…

Sometime in late November, I was on the phone stuck in a virtual queue while idly skimming through the AI news of the day, when my eye was drawn to some content about Google DeepMind’s new Gemini 3 AI’s ability to ‘vibe code’.

TavernQuest? PubWars? BeerHunter?

It’s probably relevant at this point to mention that I am a huge fan of tabletop and video games. A while ago, my kids and I had made up some alternative rules to an old boardgame called Heroquest. Our rules were of the ‘cozy’ variety and they enabled the player to run a ‘friendly’ cafe for the questing heroes and dungeon-dwelling monsters alike.

While being forced to listen to some awful hold music, I absent-mindedly dropped our homebrewed boardgame rules – which mostly consisted of a spreadsheet of tables and lists – into Google’s AI Studio and challenged Gemini 3 to ‘make this game’*.

And it did. Impressively well.

With that, my interest was piqued, a creative seed was planted, and I jumped into the rabbit-hole with both feet.

(*I’ve romanticised this somewhat – the initial prompt was a bit more complex than stated, as I had previously done some work with other AI suites to expand the idea as a standalone boardgame or a Roblox game.)

Building a Prototype
Image generated from game code by Google NotebookLM’s amazing Infographic feature

What I’ve made looks and sounds like something from the dawn of videogame development, but the rudimentary graphics and bleepy sound effects serve their purpose to showcase the ideas and gameplay loops that I’ve developed.

Tavernkeeper is an unashamed homage to games of my childhood; what if the characters from Gauntlet and Golden Axe took a breath from their dungeon-delving and met up for a cold beverage in Tapper’s bar? What happens when they rub shoulders with an off-duty goblin gang from the next dungeon? Or the zombified local farmer still sticks to old drinking habits? Is it a hygiene concern if I hire a Mummy as a cook?

I know, it’s got looks only its creator could love.

It was an incredibly rewarding experience to be able to start with a simple boardgame grid and a turn-based game framework then introduce customers to test by tweaking behaviours and interactions on the fly – all with no tangible coding knowledge. Working with Google’s AI Studio code assistant, I could apply the iterative improvements as the ‘human in the loop’ by quickly playtesting the existing game and providing feedback and observations to modify and expand the game features.

What started off as a basic boardgame-friendly dungeon room ‘side quest’, gradually evolved into a very silly, complex simulation of a tavern full of fantasy tropes. Locals visit in hope of assistance with their missing cat/haunted well/unwanted houseguest, heroes gather to plan quests while dread knights move to foil them, necromancers recruit minions, vampires and assassins stalk their prey, there are brawls and bawdiness, secrets and ghosts.

Gamebox concept art by Gemini
The Ghost in the Machine
Bernard the Ghost is currently more grumpy than helpful.

Watching the ‘ghost in the machine’ convert my ideas into code at lightning pace genuinely left me in awe. I would often outsource some of the bigger ideas to another AI tool to produce content – for example the myriad phrases (‘barks’) uttered by the many game characters were largely generated by Anthropic’s Claude based on a few flavour examples.

I’ve even used a live AI link (Gemini 3 Flash API) to provide ‘context-aware narrative experiences’. The Spirit Mirror is essentially an AI chatbot acting the part of a ghostly former tavernkeeper who can provide storyline hints and gameplay guidance, each shift is summarised by an AI-generated Morning Gazette article that appraises the player’s performance, and an obituary is written for the player when they inevitably get done in by an angry customer.

Coding Heresy or The Future of Game Development?
Maybe it should stay as a board game?

I am anxiously aware of the fact that this whole concept of a non-programmer like me ‘vibe coding’ is probably offensive to the entire software development sector; just as the clinical sector I’m more familiar with is very wary of any AI encroachment on patient care and data, and the writer in me remains repulsed by the idea of auto-generating written content (to assuage my guilt and address my hypocrisy, I am definitely planning to go through the barks and edit/replace many of them).

However, the absolute joy that AI has given me to be able to see ideas come to life in minutes has been a powerful and addictive timesink. I wish I could do this for a living.

The ability of AI tools to facilitate the realisation of ideas at pace is exciting and I’m sure can be applied to many industries.

Now I’m torn as to whether to keep exploring creative avenues or accepting my fate as a healthcare professional for life. I wonder where the videogame industry stands with AI use? Would I be able to crowdfund some proper development of this idea, with human-led art, sound, and design?

Or could I go all the way to market with a product entirely built with AI-generated content? Game designers must be using these tools now, right? Where is the line?

Phew, AI image gen still can’t spell. Breathe easy fellow writers.

2 Comments

  1. Anonymous Anonymous

    This looks really cool

    • Mathew James Westhorpe Mathew James Westhorpe

      Thanks, I appreciate the positive feedback.

Leave a Reply