Intro
Inspired by Curiosity Traps, I wanted a curiosity honeypot on my blog.
Obviously by making a post about it, it’s not that hidden. See if you can find it.
In 2026 with AI, it’s easier than ever to create vibe-coded shovelware.
But I’ve always wanted to make games. It’s just I never had the time. I didn’t have graphics abilities or the equipment to efficiently create music.
That’s okay. The game doesn’t need to be that great. I can be the creative director and let Claude do the technical heavy lifting.
Clippy2027
Level 1: a tapping game of permission prompts
Level 2: bending wire by hand
Level 3: an idle clicker economy
Level 4: dialogue with skill checks
Level 5: a factory builder
Level 6: an annoying dating sim
Level 7: real-time planet capture
Level 8: a turn-based word battleInspiration
Clippy2027 is a very derivative game. It borrows from Factorio, Disco Elysium, Galcon 2 and Undertale. A lot of old ideas shoved into a new theme.
I wanted to draw from AI 2027, an AI fan fiction page that tries to predict the future.
I wanted to force the game player to think a little bit about what they’re doing, and what the future entails if AI is allowed to do whatever we ask it to do.
It’s also a nod to the Universal Paperclips story and game.
The levels are short. Some endings are easy. Other endings are challenging. Choices matter. There’s lots of AI-written dialogue, but hopefully it’s fun.
Any feedback is welcome.
Tech
I really like ink, which is a library for doing dialogue trees and compiling them down.
I sort of like the audio library I used, Tone.js. It doesn’t sound that great, but it does give me the flexibility to subtly alter how the sound works as the game ramps up or down.
Really good games have really excellent sound. There are really subtle transitions in the music as things get more dramatic. I wanted to replicate that in a lightweight way.
Otherwise, it’s your run-of-the-mill JavaScript clicker, but perhaps this will be a gateway game for me to build more complex games.
Working with Claude
One aspect of working with Claude that made this game very easy to make was the fact that we could build a simulator.
Really great games require hours and hours of test playing to make sure it’s balanced and works well.
You can’t fully replace test playing, but you can use a simulator to see how changes impact how difficult something might be.
For example, how long should it take to do something like make five paperclips? If you tweak a difficulty on something, on average, how often does a player lose?
Taking the human out of the loop here does make it harder to tell how difficult something is subjectively. But if the simulator is making the same decisions, you have an anchor of objectivity so that you can see the effect of something changing. That’s harder to do with human test playing.
And of course Claude is just really fast. For art direction, I can ask for a huge swath of ideas. It’ll come back in a few seconds and I can pick a favorite. Claude can whip up an audio sampler for me, again in seconds. For a game in this format, it’s very easy for me and Claude to collaborate quickly.
Choices matter
One of my favorite tags on Steam is Choices Matter.
I just like it. There’s of course dread when I make a choice knowing that I’m blocking off a tree of things that I won’t be able to do.
In games like Mass Effect, I can’t stop myself from looking at the wiki to see what would happen if I made a different choice. Sometimes I save scum and satiate my curiosity by exploring different routes.
And of course the pinnacle of this is Disco Elysium. So many choices to make that impact you as a character and of course the events of the story.
In Clippy2027 I wanted some aspects of this.
For example, in the very first level it records how many times you allow and generally how resistant you are to automation.
In subsequent levels when it’s trying to convince you of something, if you were previously very resistant to automation, it tries to convince you to automate further.
If previously you were very eager to automate away and allow Clippy to make decisions, it does the opposite and tries to argue that things are spiraling out of control.
The simulator helps me see the effects of the dialogue difficulty in different camps based on previous decisions.
Conclusion
It’s just a silly game. Try to find it. Explore some endings. See if you can beat it. And then think about what will happen in REAL LIFE when we start giving our super powerful AIs goals.