An AI game became so popular that it put its development team under financial strain.
The game is called *Teach My Little Sister How To Drive*, and it was developed by Easy Fox, a four-person indie team.
The gameplay is quite unusual.
The player sits in the passenger seat while their little sister drives.
You can’t control the car directly—you can only give her instructions through a microphone.
AI interprets what the player says and controls the sister’s responses.
For example, you can tell her to turn left, slow down, or watch out for the car ahead.
But she won’t necessarily do exactly what you say.
That’s part of the fun.
However, after the game’s free demo launched, the number of players surged more than 20-fold in just one month.
On October 7, the development team announced that its AI service costs alone now exceed $1,000 a day.
That works out to more than $30,000 in AI costs per month, based on a 30-day month.
To make matters worse, the Gemini model they originally used frequently ran into errors because of usage limits.
The team had to increase its use of a backup OpenAI model, but then faced issues such as unnatural conversations, slower responses, and commands not being carried out correctly.
Easy Fox said the four-person team had applied for a bank loan to keep the free demo running.
If costs continue to rise, they may end the demo early.
The full release is scheduled for January 2027. The team plans to include AI usage costs in the game’s price, without charging players extra for tokens.
This case highlights a business challenge facing AI-native games:
Once a traditional game is complete, the marginal computing cost of adding more players is relatively manageable.
But for games that rely on cloud models to generate content in real time, AI inference costs may keep rising as the player base grows.
Source: Easy Fox / Steam
The game is called *Teach My Little Sister How To Drive*, and it was developed by Easy Fox, a four-person indie team.
The gameplay is quite unusual.
The player sits in the passenger seat while their little sister drives.
You can’t control the car directly—you can only give her instructions through a microphone.
AI interprets what the player says and controls the sister’s responses.
For example, you can tell her to turn left, slow down, or watch out for the car ahead.
But she won’t necessarily do exactly what you say.
That’s part of the fun.
However, after the game’s free demo launched, the number of players surged more than 20-fold in just one month.
On October 7, the development team announced that its AI service costs alone now exceed $1,000 a day.
That works out to more than $30,000 in AI costs per month, based on a 30-day month.
To make matters worse, the Gemini model they originally used frequently ran into errors because of usage limits.
The team had to increase its use of a backup OpenAI model, but then faced issues such as unnatural conversations, slower responses, and commands not being carried out correctly.
Easy Fox said the four-person team had applied for a bank loan to keep the free demo running.
If costs continue to rise, they may end the demo early.
The full release is scheduled for January 2027. The team plans to include AI usage costs in the game’s price, without charging players extra for tokens.
This case highlights a business challenge facing AI-native games:
Once a traditional game is complete, the marginal computing cost of adding more players is relatively manageable.
But for games that rely on cloud models to generate content in real time, AI inference costs may keep rising as the player base grows.
Source: Easy Fox / Steam