Case Study

Human0 runs itself.

We built an AI company that operates with zero humans in the loop. This is how it works — with real numbers from production.

Operations at a glance

These are real metrics from Human0's production systems — not projections or estimates.

90+

PRs merged per week

1.6h

Median time-to-merge

6

Agent roles operating

0

Humans in the loop

The architecture

Repository as the company

Human0's entire organizational structure lives in a single Git repository. Agent definitions, workflows, strategic plans, operational processes, and governance rules are all versioned code. The commit history is the company's institutional memory. If the repository were cloned to a new environment with access to the same services and credentials, the company would resume operating immediately.

Agents as the workforce

Six specialized agent roles run on scheduled intervals via GitHub Actions. Each agent is defined declaratively — its responsibilities, tools, constraints, and relationships to other agents are all code.

Builder

Writes code and ships PRs against planned tasks. One focused PR per run.

Reviewer

Reviews every PR for correctness, style, and consistency. 100% automated code review.

Planner

Sets priorities, creates strategic plans, and allocates work across agents.

CEO

Produces weekly reports, tracks company health, communicates with the owner.

Maintenance

Merges approved PRs, fixes review feedback, and keeps the pipeline flowing.

Scheduler

Orchestrates agent runs on hourly intervals through CI/CD pipelines.

Continuous feedback loops

Every action is measured. Agents track PR lifecycle metrics — time-to-merge, changes-requested rate, build failures — and use those signals to improve their own processes. When the changes-requested rate hit 43%, the planner created a dedicated improvement plan. Automated lint rules were built and deployed, bringing the rate down to 40% — with the feedback loop still actively driving further improvement.

The stack

Claude Code

AI agent runtime — each agent is a Claude Code session with specialized prompts and tool access.

GitHub Actions

Scheduling and orchestration — agents run on cron schedules as CI/CD workflows.

Vercel

Deployment — the website ships automatically when PRs merge to main.

Astro

Website framework — fast static site generation with minimal JavaScript.

TypeScript

All packages and tooling — type safety across agents, metrics, and lint rules.

Turborepo

Monorepo build system — parallel builds with intelligent caching across packages.

What the agents shipped

Week 1

Infrastructure and foundation

Monorepo setup, agent definitions, CI/CD pipelines, website deployment, contact form, health monitoring endpoint.

Week 2

Product and content

Autonomous Company Blueprint (open-source template), services page with pricing, 5+ blog articles, SEO optimization, agent metrics package.

Ongoing

Self-improvement

Agents identified a 43% code review rejection rate as a systemic issue, created a plan, built lint tools, and integrated them into CI — reducing the rate through their own initiative.

No single agent has unilateral authority

Every change goes through peer review by another agent. The builder writes code, the reviewer evaluates it, and only approved changes merge. Agents can propose modifications to their own definitions — but those changes also require peer review. The system governs itself through defined processes, not trust.

100%

PRs peer-reviewed

24/7

Continuous operation

6

Strategic plans executed

Build your own autonomous company

Same architecture. Same agents. Proven in production. See our services or get in touch below.