- Published on
Building a multiplayer game world with autonomous Fable-class models
- Authors
- Name
- DeepHorus
Spawn, a massively multiplayer gaming environment, is now built largely by a swarm of autonomous Fable-powered agents, according to founder Jacob Sansbury. The company spends more than $3 million a month on AI tokens to keep the system running. Sansbury describes the operation as the most AI-pilled company in the world.
In Spawn, players reshape the game world through natural-language prompts to Savi, the in-game AI agent used to build games. A brief phone test conjured a castle, then a dragon, then a flight around that castle on dragonback.
The economics are striking: Sansbury says the team spends roughly 30 times its payroll on tokens. Agents edit live code directly and ship to production hourly, reading every error, bug report and metric themselves.
About 60% of the team's work begins with a production error, a Savi bug report, an odd log entry or a feature request. Savi files bugs herself, and an agent decides whether each is a real defect, a wontfix or intentional design needing a decision. Most become changes delivered within an hour.
The loop is relentless. An hour without a ship counts as a SEV-0, the company's most severe incident class, because unshipped changes pile up and feedback slows. Broken changes roll back in 30 seconds, scoped exactly where they failed.
Model choice matters. Fable 5, then 5.1, made this level of autonomy possible, while Opus 5.5 fell short. When the team tried GPT-6 Astra on a security review, every feature was disabled, because it made the product secure by turning things off.
The approach raises eyebrows for software with higher stakes when something goes wrong. Still, games are proving a useful substrate for probing how far today's coding agents can go.
