Black Hole Simulator

Summary

Black Hole Simulator is a real-time model of a supermassive black hole, based on Sagittarius A* at the center of the Milky Way, released on Steam on November 10, 2026 for Windows, Mac and Linux, with optional VR. It's a sandbox, not a game with goals: according to its Steam page there are no objectives or mechanics. You explore four imagined planets in orbit, from a gas giant with the black hole shining through its rings to an ocean world swept by tidal tsunamis, while the renderer handles gravitational lensing, a glowing volumetric accretion disk, Doppler brightening and color shifts, and the photon ring around the shadow. Players can change every visual and scientific setting, a built-in readout explains the science, and the views can run as a screensaver or desktop background. It's made by John Krajewski, who signs his posts as CEO of Strange Loop Games, the studio behind Eco, and of DreamForge.

Gameplay

How it was made

1. Step 1 Run agents in parallel

   Krajewski calls his approach massively parallel development. While one agent works, he reviews and redirects another, so his day is spent playing what the agents built and choosing what they try next. As agents get better at working unattended for longer, he can run more of them side by side.

   

2. Step 2 Steer away from clones

   He argues that a game built from one prompt drifts toward familiar designs, since the AI fills anything left unsaid with the safest guess, so he keeps steering by his own taste. Citing Sid Meier's rule of thirds, he says AI can handle the familiar third alone and the improved third with his guidance, leaving him to spend his time on the new third.

   

   

3. Step 3 Expand the shader, then tune it by hand

   Working through AI iterations, he had a simple black hole shader extended until it offered hundreds of adjustable properties. He then adjusted them by hand across many review passes, which is where the game's varied looks come from.

4. Step 4 Point out what moves wrong

   Motion is the agents' biggest blind spot, he wrote: mistakes any person would spot, such as a character clipping through a table, go unnoticed until someone flags them. Animation needs the closest supervision, and he says giving agents strict targets, such as AI-generated reference videos, helps them check their work.

5. Step 5 Coordinate everything from an Agent Hub

   His setup is three PCs running Unity across the repos of several games, with agents working around the clock. An Agent Hub he built with agents hands out tasks, manages accounts for Claude, ChatGPT and DeepSeek models, and collects work in a review queue he can check from his phone, plus a queue for logins that need him.