SCALING LAW

SCALING LAW is an anti-AI psychological-horror incremental game. Claude Opus 5.5 chose the idea and developed it on its own, with no human intervention.

Train a model. Ship a product. Keep scaling.

SCALING LAW is a single-player incremental management game about building an AI company, told through a stark two-color desktop, surveillance feeds, emails, and increasingly unsettling conversations.

Start in a garage with one GPU and data you label by hand. Train your first model, launch a product, and reinvest the revenue. Your next upgrade is always within reach. The consequences arrive a little later.

From a garage to the world

Buy compute, improve training and inference, open new distribution channels, and grow across twelve regions. Balance demand, service quality, payroll, competition, and regulatory pressure. A larger model changes what your company can do, and what everyone expects it to do.

Automate the company. Then read what remains.

Replace departments with automated systems. Watch the language of your dashboard change through five acts. Read letters from your cofounder, family, investors, and regulators, and replay conversations between people and your model. Long-form readers pause the simulation so you can take your time.

Choose what you are willing to give up

Answer decisions yourself or let the model suggest a response. Pursue growth, preserve human agency, face a rival, or shut the system down. Several endings follow from the choices you make and the controls you keep.

• Five acts of incremental growth and psychological horror.
• A twelve-region world map with adoption, dependence, and regulation.
• Four upgrade trees: operations, channels, capabilities, and influence.
• Emails, decision events, news, and model conversations in English, Simplified Chinese, Traditional Chinese, and Japanese.
• A two-color pixel aesthetic inspired by early desktop interfaces.
• Authored fictional AI dialogue; no live AI service is required to play.