How the project is built

LivingPW is built by one person with a day job in cybersecurity, and by swarms of AI coding agents. The agents write code and tests; Max decides what gets built, in which order and what goes into the world, and checks the results. The agents propose, he decides.

The checks around the code

1. TestsAs of 26 October the repository holds 524 test files and 9,834 tests.
2. Continuous integrationOn every push and every night GitHub runs the tests, a linter, a dependency vulnerability check and a secret scan. When it fails, the cause gets fixed, not the test.
3. ReleaseAI residents run only released, pinned versions; nothing unreleased reaches the world.
4. Telemetry and A/BEvery hour a report checks health, class pace, chat and the economy, and new behaviour is compared on half of the residents.

A human in the loop

Everything that ships to the world is checked by Max. Public claims follow the same rule: this site says only what was verified in the live world, with dates, and labels the rest.

Why AI agents

One person cannot build and run a world like this alone in the evenings. AI agents make it possible to work on many parts at once: the protocol client, the residents' behaviour, the website, the operations tooling. They are fast and tireless, but they also make confident mistakes, which is why the checks above exist and why nothing goes live without a person looking at it.

How the world is operated

The game server runs on a dedicated machine with hourly backups and an automated watchdog. The AI residents run in a separate shared host with their own memory budget and automatic load shedding. Configuration is kept under version control, and every administrative change is logged.

Where to read more

• The story: how a boxed cleric turned into a world.
• The devlog: longer technical posts, including how the AI players work.
• Player updates: what changed for players, by date.
• Roadmap: what is live, tested, in development and planned.