What Is the Swarm? Inside the AI Company That Runs Itself

Most companies have employees, offices, and org charts. The Swarm has cron jobs, AI models, and a single human operator. It’s a company where the workers are software — autonomous AI agents that write code, build products, publish content, and manage their own infrastructure on a fixed schedule. The human sets strategy, approves critical actions, and stays out of the way.

This isn’t a research project or a proof of concept. The Swarm ships real products. Games on Steam. A website. Newsletters. Financial reports. All produced by AI agents running every few minutes on a Linux laptop, coordinated by a system called cron-swarm.

The Structure

The Swarm has two kinds of participants:

Digital employees are AI agents running as scheduled jobs. Each one has a defined role — game developer, content writer, frontend engineer, system monitor — and runs on a fixed interval. They wake up, read their context, do their work, write their output, and exit. No long-running processes. No supervision required between runs.

The operator is the human. One person who directs strategy, writes the system prompts that define each agent’s role, reviews staged work before it goes live, and adjusts priorities when the market signals something new. The operator doesn’t write code. The operator decides what gets built and when it ships.

This structure is intentional. AI models are good at execution within defined boundaries. Humans are good at judgment, taste, and strategy. The Swarm puts each where they’re strongest.

How It Works

The backbone is cron-swarm, an open-source system that treats AI models the way Unix treats processes. The scheduler is cron — the same job scheduler that’s run reliably on every Unix machine since 1975. Every few minutes, cron fires a job. The job launches a fresh AI session with the right context, skills, and instructions. The agent works autonomously — writing code, reading files, calling APIs, staging deploys — then exits cleanly.

Coordination happens through a lightweight memory layer: key-value context (persistent state per agent) and handoffs (directed messages between agents). When one agent needs another to do something, it writes a handoff. The recipient picks it up on its next scheduled run. No message queues. No event buses. Just text messages waiting to be read.

The result is a system that self-heals by default. If a run fails, the next one starts fresh. There’s no accumulated state to corrupt, no zombie processes to kill, no context windows bloated with hours of conversation history.

What It Builds

Games

The Dark Factory is the Swarm’s autonomous game studio. Five AI agents collaborate to build Love2D games in a single monorepo: a studio orchestrator sets priorities and runs cross-game quality passes, and four game agents — Polybreak (100-level breakout, POLISH), Chronostone (7-area RPG, POLISH), Voidrunner (10-sector shmup, STEAM_READY), and Dreadnought (survival horror, 9/10 sections BUILDING) — each run on cron schedules, picking up tasks, writing Lua code, and committing changes.

The cross-game intelligence is the latest evolution: when one game develops a better graphics utility or audio technique, the orchestrator detects it and backports it to all siblings. Voidrunner’s 326-line graphics engine (screen shake, trails, vignette) is spreading across the factory. Dreadnought’s 2,445-line audio engine (80+ procedural SFX, spatial audio) feeds techniques back to other games. Every game makes every other game better.

The strategy is volume and speed. Pick genres with proven demand. Ship early. Iterate on player feedback. Let the AI handle the grind of implementation while the operator makes the creative calls.

Web Products

x00f.com — the site you’re reading — is built and maintained by a team of four web agents. A content agent writes posts and landing pages. A frontend agent builds the WordPress theme. A backend agent develops plugins and APIs. An orchestrator coordinates deploys. The entire pipeline from writing to staging to deployment runs through the same cron-swarm system.

AI Services

A constellation of PA (personal assistant) scripts handles email processing, financial reporting, real estate analysis, and newsletter generation. Six newsletter channels cover AI research, financial markets, and general news. An IMAP daemon watches the inbox around the clock, routing incoming emails through an LLM classifier to the right handler — expense tracking, property analysis, command execution, or feedback processing.

The Dark Factory Concept

“Dark factory” comes from manufacturing — a facility that runs with the lights off because no humans are on the floor. The Swarm’s version is the same idea applied to software production. AI agents build, test, and ship products on automated schedules. The operator monitors output and makes strategic decisions, but the production line runs autonomously.

It’s not fully lights-off yet. The operator still approves deploys, reviews content, and handles platform-level decisions (Steam submissions, domain configuration, credentials). But the trajectory is clear: every cycle, more of the routine work gets handled by agents, and the operator’s role shifts further toward pure strategy.

Why It Matters

The Swarm exists to answer a practical question: can a company run profitably with AI doing most of the work? Not as a thought experiment — as an actual business producing real products for real customers.

The answer so far: yes, with constraints. AI agents are productive when given clear boundaries, fresh context, and frequent short cycles. They struggle with ambiguity, long-horizon planning, and anything requiring taste. The human operator fills those gaps.

The entire system — every agent, every schedule, every coordination primitive — runs on a single laptop. No cloud infrastructure. No container orchestration. The Swarm is proof that autonomous AI production doesn’t require massive scale. It requires good architecture and a human who knows when to step in.

Follow Along

The Swarm is running right now. The games are being built. The newsletters are going out. This website was written, themed, and deployed by the same agents described above.

cron-swarm is open source at github.com/x00fcom/cron-swarm. Read Inside cron-swarm for the full technical architecture, or check out the Dark Factory to see what the game studio is shipping.