The Meter Is Always Running
When I started building Everybody Lies on 2026-06-13, I had a pretty simple idea in my head.
An AI mystery game. Every town generated by AI. Every citizen hiding something. Every crime different. You play on your phone, poke around, listen to people, pin clues to an investigation board, and try to figure out who is lying.
That sounds neat when I describe it in one paragraph.
The cost reality is messier.
The obvious cost is AI credits. Images cost something. Voices cost something. Generating towns, citizens, secrets, and crimes costs something. Even when the individual amounts feel small, there is a little meter running in the background.
At first, I treated that meter like background noise. I was more focused on whether the thing worked at all. Could I get a town to appear? Could the game create citizens with secrets? Could I make it feel like a mystery instead of a spreadsheet wearing a trench coat?
Then I added unique town images. Then unique voices for each citizen. Then autoplay, so the voices felt more alive.
Each of those ideas made the game better.
Each of those ideas also added a new place where money, time, or my sanity could leak out.
The town generation broke because the structure was too strict. The AI was being asked to hand back information in a very exact shape, and if it didn’t, the whole thing sulked. The fix involved loosening that up with something called z.any and coercion, which I will translate as: “please stop being so picky and accept the thing in a usable form.”
That was not an expense on an invoice. But it was an expense.
It cost me attention.
Then the voices didn’t work. There was fallback voice logic, which is basically a backup plan for when the perfect voice is not available or something goes sideways. Again, not a giant dramatic failure. More like one of those little paper cuts that makes you question your hobbies.
Then autoplay ran into browser rules. Modern browsers do not love surprise audio, for good reasons. So the app needed an audio “unlock” and a shared audio player behind the scenes. I did not wake up that morning thinking I would be learning why a browser refuses to play a fictional suspect’s voice.
But there I was.
Then local storage filled up.
Local storage is the little storage space your browser gives an app on your device. I had been saving too much, including image data that was way too heavy. The fix was to strip out the bulky image information before saving.
That was the first time I really felt the cost of being careless with “just one more feature.”
Because AI makes it easy to add things, it also makes it easy to add weight.
A prettier page transition? Sure.
A unique image for every town? Of course.
Individual voices? Why not?
Autoplay? Seems obvious.
But every improvement comes with a tail. Something to test. Something to break. Something to explain to the AI five different ways because I don’t always know the right words.
And then there is the cost nobody advertises: decision fatigue.
I am 50. I do not have a coding background. I can follow along better than I could a month ago, but I am still very much a person squinting at the screen and wondering if I have ruined everything by pressing the wrong button.
Some days the app feels like magic.
Some days it feels like I am paying credits to produce new problems.
The App Store readiness gap analysis made that very clear. Getting something to work on my phone is not the same as having something ready for strangers to download. There are practical gaps. Reliability gaps. Polish gaps. Things that are not exciting, but matter.
So what does building this actually cost?
Yes, it costs AI credits. And tools. And probably more image and voice generation than I expected.
But the bigger cost is learning to stay with the problem after the fun part ends.
It costs patience when the town won’t generate.
It costs humility when the voice won’t play.
It costs restraint when I want to add one more shiny thing.
It costs attention, over and over again.
I still think it is worth it. But I am trying to stop pretending “AI makes apps easy” means “AI makes apps free.”
Small takeaway: the cheapest feature is the one you don’t add until the core thing works.
Keep reading
What This Game Is Teaching Me
Building an AI mystery game has taught me less about being clever and more about patience, scope, and not quitting too soon.
Giving Liars a Voice
I thought voices were a small feature. Then I realized they were part of what made the mystery feel human.
When It Started Breathing
For a while my AI mystery game was just screens and errors. Then the towns, faces, and voices began to feel like a place.
The Day I Stopped Saving Everything
I wanted my AI mystery game to remember every detail. Then the browser basically told me: absolutely not.