Show HN: A Company Running in an AI Harness
An autonomous operating system where AI agents manage routine workflows and business operations.
1Autonomous Agency Architecture
The platform functions as a persistent Linux environment where AI agents handle routine business tasks. By integrating Git, browsers, and desktop tools, it allows agents to coordinate workflows independently. This system enables a single founder to manage complex projects by delegating operational chores to specialized AI workers while retaining final judgment.
2Agent Roles and Responsibilities
Different agents are assigned specific roles such as creative direction, build management, and research. For example, agents like Rosa and Jun handle artistic and puzzle-related tasks, while others manage playtest invitations or feedback sorting. This structure ensures that routine production work continues overnight, allowing human leads to focus solely on high-level decisions.
3Human-in-the-Loop Decision Making
The system enforces a strict boundary between automated execution and human oversight. While agents manage builds, exports, and bug fixes, critical milestones like approving a playable demo require human intervention. This ensures that the company maintains its core vision and quality standards without the founder needing to micromanage every individual task.
4Operational Workflow Integration
Workflows are managed through briefs that define company priorities and goals. When a task is initiated, the system tracks progress across multiple devices and platforms. If an agent encounters a problem, it can restart sessions or swap models to maintain momentum, ensuring that the project scope remains small and manageable throughout.
5Security and Authority Controls
The platform includes built-in safeguards to prevent unauthorized actions. Agents cannot grant themselves new authority or approve expansion without human consent. Financial budgets and credentials are kept outside the primary company computer, ensuring that the AI operates within defined parameters while maintaining a clear record of every external action taken.
6Consistency and Quality Management
The system maintains a unified voice and visual language across all outputs. If a core fact or design element is updated, the platform automatically flags any previous work that relied on the outdated information. Users can also set quality tiers, such as Fast or Frontier, to dictate how deeply agents investigate tasks.
7Technical Implementation and Deployment
The software is designed for flexibility, supporting macOS, Linux, and Windows with WSL2 via Docker. Users can choose between a self-hosted version for unlimited companies or a managed hosted service. The hosted plans provide varying levels of AI usage credits, priority support, and larger computing resources for scaling business operations effectively.
8Future of Automated Management
By treating a company as a persistent computer program, this tool shifts the focus from manual labor to strategic oversight. It demonstrates how AI can absorb routine operational friction, allowing small teams to achieve output levels previously reserved for larger organizations. The platform is currently in development preview and open source.
Original article: Hacker News: Show HN — Show HN: A Company Running in an AI Harness