Creating BY-ELECTION!
https://alexandergriffiths.itch.io/by-election
What is the game?
A free game I released on 20th July 2026 on itch.io. The game is a simulator where you are the candidate running in a British parliamentary by-election.
The game - https://alexandergriffiths.itch.io/by-election
Why a by-election simulator?
I thought it would be fun.
No one else had done it.
It’s small and self-contained meaning I thought there was a chance I could make it work.
A few years ago (before the advent of LLMs) I tried to create and code the game myself, and with no coding skills made a basic notepad based version game that was rubbish (and before you ask which I cannot find).
By-elections are having something of a moment in 2026 UK politics and so I thought it might capture the mood.
I wanted to create something that could be both political and fun for all. So much of British politics right now is dark, depressing and contentious and this game, which eschews heavy policy for lighter jollity, is trying to bring a little sunshine into the lives of those burdened with an interest in British politics.
What was the creation process like?
Easy and hard.
Easy because with any large text blocks Claude had enough context both from me and from its corpus of knowledge to fill in the blanks when I didn’t elucidate fully. As stated below with text especially it was a reliable partner and when it worked it really gave you the confidence and enthusiasm to keep going.
However, when problems did emerge it wasn’t merely that it would take a while to fix, it was that it could actively go backwards and make things worse or use up my usage whilst repeatedly being unable to complete the task. What was also worrying was that when it was struggling it would often simply fail and repeat without explaining to me any of the challenges it faced and it took me a while to work this out to my cost.
The biggest challenge though was my own ignorance of what would and wouldn’t be the biggest challenges for the game. Naively I thought the game mechanics would be the most difficult to programme when in reality these worked easily and without any (touch wood) bugs emerging.
What worked?
Text and mechanics.
The implementation of the game mechanics and text creation was shockingly smooth. Whether it was creating new event chains, deleting two weeks out of the game or changing the value and effect of different decisions the process of adding new text and mechanics worked seamlessly. Reviewing the text and the specific mechanics I made a good number of edits and changes reflecting my own preferences but was overall satisfied with the process.
What didn’t work?
Movement and common sense.
Movement was the biggest challenge to my ideas and the only one where the game never got even close to what I wanted. However I realised after I finished the game that this was in great part due to the problem of having to rely on Opus for most of the process because Fable was banned by the Trump administration. However, even with Fable a number of issues remained and so movement, save a couple of simple examples like the colour change rosette and the loading bar, I left this area alone but it was the one area I found Claude to be pretty disappointing and I’d consider looking elsewhere next time.
Common sense however was something that was quite a surprise to find in short supply. Being a one man game designer I had no partner to look over my shoulder and critique my ideas and so for much of the design process I had the by-election causing MP given the (seemingly) generic name of John Smith because I wasn’t paying enough attention. However, as many will know there is a particularly notable John Smith MP who was leader of the Labour Party in the 90s who died of a heart attack in 1994. Using the name John Smith therefore was tactless and should have been a red flag Claude picked up but it was left to me to notice and correct this (thankfully well before publication) and a number of other minor errors which could have courted controversy. For the rest of the design process therefore I found myself somewhat paranoid, checking and rechecking to see for any similar mistakes and (hopefully) none have been made.
How much did the game cost to make?
Looking back through my billing, in addition to the usual monthly fee I pay to Anthropic I can see that I spent well over £100 on usage credits which considering it took me just under a month to create the game means I could have upgraded from the Pro plan (currently advertised at £15/month) to Max (currently advertised at £90/month) and probably saved money.
I am not particularly pained by this fact though because the whole creation process was a journey of discovery and I did not know at the outset how much I’d use or where the pain points were, or where the places where the greatest usage would be.
What did you learn from the project?
Game creation with Claude does work BUT you need to use Fable 5 or have incredible amounts of patience with Opus. If it hadn’t been for the Trump administration shutting off Fable access for part of the design window the game would have been finished much faster. If I was to create another game using Claude, just seeing the difference in outcomes between Opus and Fable means I would never consider beginning the project unless I was using Fable.
More broadly though, having had a first experience of AI assisted game creation I would be more willing to shop around other AIs next time for what I would want to achieve now I have some idea of the components and process. That said whilst I’m sure there are other AI that could have done it faster, smoother and more easily I overall found Claude a good partner on this project.
What do you think are the wider implications of AI assisted game creation?
Well I think the really interesting question is not about games but other creative endeavours you could use AI for. In addition to game creation I have long harboured the idea of one day making my own film and with AI I think it’s possible. The core challenges to this, and which would stay my hand at present are not technological but legal (in this game I avoided using real political parties and fiction well out of copyright ahem hint hint) and experiential, the challenges would obviously be bigger and require different skills to those I’ve developed here.
I doubt very much though that Claude is the best AI for trying to create my own film though and so before I even start trying that option I need to research what’s on offer first.
Did you actually write this post or did you just prompt Claude?
I wrote this post myself and only ran it through Claude for proofreading. I did not want Claude to write this because this is about my own assessment of the game process and my prompts only occasionally included my thoughts on the process so I don’t believe Claude would be capable of giving actual insight into what I thought because so often I kept it to myself.